CoChat
Sign Up
Cited by 2,058Medical Image Segmentation TechniquesArtificial Intelligence in Healthcare and EducationAdvanced MRI Techniques and Applications

An overview of deep learning in medical imaging focusing on MRI

Open access
Lundervold, Alexander Selvikvåg, Lundervold, Arvid·2019·Zeitschrift für Medizinische Physik
DOI 10.1016/j.zemedi.2018.11.002
Open this paper in CoChat — free account

Abstract

What has happened in machine learning lately, and what does it mean for the future of medical image analysis? Machine learning has witnessed a tremendous amount of attention over the last few years. The current boom started around 2009 when so-called deep artificial neural networks began outperforming other established models on a number of important benchmarks. Deep neural networks are now the state-of-the-art machine learning models across a variety of areas, from image analysis to natural language processing, and widely deployed in academia and industry. These developments have a huge potential for medical imaging technology, medical data analysis, medical diagnostics and healthcare in general, slowly being realized. We provide a short overview of recent advances and some associated challenges in machine learning applied to medical image processing and image analysis. As this has become a very broad and fast expanding field we will not survey the entire landscape of applications, but put particular focus on deep learning in MRI. Our aim is threefold: (i) give a brief introduction to deep learning with pointers to core references; (ii) indicate how deep learning has been applied to the entire MRI processing chain, from acquisition to image retrieval, from segmentation to disease prediction; (iii) provide a starting point for people interested in experimenting and perhaps contributing to the field of deep learning for medical imaging by pointing out good educational resources, state-of-the-art open-source code, and interesting sources of data and problems related medical imaging.

AI snapshot

Written from the paper’s full text.

TL;DR

This narrative review surveys deep learning in medical imaging with a focus on MRI, covering the MRI processing chain from acquisition and reconstruction to segmentation, diagnosis, and image retrieval. It argues that deep learning is already state-of-the-art in many imaging tasks but faces major challenges in data access, interpretability, workflow integration, and regulation.

Key findings

  • Deep learning, especially CNNs, has become the de facto standard for many computer vision and medical image analysis tasks, and is being applied across the entire MRI workflow: acquisition/reconstruction, QSM and MR fingerprinting, denoising, super-resolution, synthesis, registration, segmentation, diagnosis/prediction, and content-based image retrieval.
  • In MR fingerprinting, Cohen et al.’s DRONE network reconstructed in 10 to 70 ms per slice, reported as 300 to 5000 times faster than conventional dictionary matching; Hoppe et al. reported matching one time series about 100 times faster than the conventional method.
  • QSMnet was trained on 60 scans from 12 healthy volunteers; DeepResolve used thin knee slices from 124 patients for training and 17 for testing, with 10 s inference time per 3D 344 × 344 × 160 volume.
  • Early deep learning for medical CBIR was disappointing: Sklan et al. reported a true positive rate of 20% on an independent test set of 2100 labeled images, though the paper states medical CBIR is now dominated by deep learning algorithms.

Methods

  • Narrative overview/review, not a systematic review or meta-analysis; the authors state three aims: introduce deep learning, indicate applications across the MRI processing chain, and point to educational resources, open-source code, data, and problems.
  • No primary data collection or new experiments; the paper synthesizes cited literature and examples across MRI acquisition, reconstruction, quantitative parameter mapping, image restoration, super-resolution, synthesis, registration, segmentation, diagnosis/prediction, and CBIR.
  • Includes short lists/tables of openly available code, medical imaging datasets/repositories, and competitions.

Limitations

  • The paper states it does not survey the entire landscape and is not comprehensive; it focuses particularly on MRI and selected organs/applications.
  • The paper states general and domain-specific challenges: lack of mathematical/theoretical underpinnings, difficulty interpreting “black box” predictions, data access/privacy issues, deep networks’ large training-data requirements, 3D CNN constraints, workflow integration, and regulatory/ethical/legal hurdles.
  • Evident from the text: as a narrative review without a systematic search protocol or quantitative synthesis, it cannot estimate effect sizes, prevalence, or comparative performance across the field.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.zemedi.2018.11.002

Cites

287 of 287 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 129 cited sources checked · 0 supported · 0 flagged

  • Findingcitation
    § 5.1. Data · p. 26

    In 3D interorgan transfer learning, pre-training a network to do brain segmentation decreased the number of annotated kidneys needed to achieve good segmentation performance.

    “One can also do interorgan transfer learning in 3D, an idea we have used for kidney segmentation, where pre-training a network to do brain segmentation decreased the number of annotated kidneys needed to achieve good segmentation performance [342].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [342] [342] A. Lundervold, A. Lundervold, J. Rørvik, Fast semisupervised segmentation of the kidneys in DCEMRI using convolutional neural networks and transfer learning, 2017. -

      Not checked in the free run.

  • Premisecitation
    § 3. Deep learning, medical imaging and MRI · p. 11

    Convolutional neural networks can reduce the gadolinium dose in contrast-enhanced brain MRI by an order of magnitude without significant reduction in image quality.

    “They can also be used to reduce the gadolinium dose in contrast-enhanced brain MRI by an order of magnitude [88] without significant reduction in image quality.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [88] [88] E. Gong, J. M. Pauly, M. Wintermark, G. Zaharchuk, Deep learning enables reduced gadolinium dose for contrast-enhanced brain MRI, Journal of magnetic resonance imaging 48 (2018) 330-340. - source

      Not checked in the free run.

  • Premisereasoning
    § 2.3. Building blocks of convolutional neural networks · p. 7

    By having each filter share the exact same weights across the whole input domain, one achieves a drastic reduction in the number of weights that need to be learned.

    “By having each filter share the exact same weights across the whole input domain, i.e. translational equivariance at each layer, one achieves a drastic reduction in the number of weights that need to be learned.”
    Quote in the paperNot a substantial claim
  • Premisecitation
    § 3.1.3. Image restoration (denoising, artifact detection) · p. 15

    Bermudez et al.'s autoencoder network outperformed the current FSL SUSAN denoising software according to peak signal-to-noise ratios.

    “Their autoencoder network outperformed the current FSL SUSAN denoising software according to peak signal-to-noise ratios.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [180] [180] C. Bermudez, A. J. Plassard, T. L. Davis, A. T. Newton, S. M. Resnick, B. A. Landman, Learning implicit brain MRI manifolds with deep learning, Proceedings of SPIE-the International Society for Optical Engineering 10574 (2018). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Premisecitation
    § 3.1.5. Image synthesis · p. 16

    Comparable tumor segmentation results can be achieved when trained on synthetic data rather than on real patient data.

    “The authors achieved comparable tumor segmentation results when trained on the synthetic data rather than on real patient data.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [213] [213] H.-C. Shin, N. A. Tenenholtz, J. K. Rogers, C. G. Schwarz, M. L. Senjem, J. L. Gunter, K. P. Andriole, M. Michalski, Medical Image Synthesis for Data Augmentation and Anonymization Using Generative Adversarial Networks, in: International Workshop on Simulation and Synthesis in Medical Imaging, Springer, pp. 1-11. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. 64
  • May combine two claims. 33

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3. Deep learning, medical imaging and MRI · p. 11

    Convolutional neural networks can be used for efficiency improvement in radiology practices through protocol determination based on short-text classification.

    “Convolutional neural networks can be used for efficiency improvement in radiology practices through protocol determination based on short-text classification [87].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [87] [87] Y. H. Lee, Efficiency improvement in a busy radiology practice: Determination of musculoskeletal magnetic resonance imaging protocol using deeplearning convolutional neural networks., Journal of digital imaging (2018). - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisereasoning
    § 2.1. Artificial neural networks · p. 4

    The change to ANNs being dominant is thanks to the growth of big data, powerful processors for parallel computations (GPUs), important tweaks to algorithms, and development of easy-to-use software frameworks.

    “This change is thanks to the growth of big data , powerful processors for parallel computations (in particular, GPUs), some important tweaks to the algorithms used to construct and train the networks, and the development of easy-to-use software frameworks.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 1. Introduction · p. 1

    A deep learning model (a convolutional neural network) halved the second best error rate on the image classification task at ILSVRC 2012.

    “Most famously on the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) 1 in 2012 [1] when a deep learning model (a convolutional neural network ) halved the second best error rate on the image classification task.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] [1] A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, in: F. Pereira, C. J. C. Burges, L. Bottou, K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 25, Curran Associates, Inc., 2012, pp. 1097-1105. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 1. Introduction · p. 1

    Deep learning rose to prominence in computer vision when neural networks started outperforming other methods on several high-profile image analysis benchmarks.

    “Deep learning rose to its prominent position in computer vision when neural networks started outperforming other methods on several high-profile image analysis benchmarks.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] [1] A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, in: F. Pereira, C. J. C. Burges, L. Bottou, K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 25, Curran Associates, Inc., 2012, pp. 1097-1105. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.1. Artificial neural networks · p. 3

    Artificial neural networks are one of the most famous machine learning models, introduced in the 1950s, and actively studied since.

    “Artificial neural networks (ANNs) is one of the most famous machine learning models, introduced already in the 1950s, and actively studied since [41, Chapter 1.2].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [41, Chapter 1.2]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 1. Introduction · p. 2

    Deep learning techniques are outperforming other approaches in natural language processing, speech recognition and synthesis, and analysis of unstructured tabular data using entity embeddings.

    “They are, however, not limited to image processing and analysis but are outperforming other approaches in areas like natural language processing [3, 4, 5], speech recognition and synthesis [6, 7] 2 , and in the analysis of unstructured, tabular-type data using entity embeddings [8, 9].”
    Quote in the paperNot a substantial claim0 of 7 sources checked
    • cites [3] [3] M. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, L. Zettlemoyer, Deep contextualized word representations, in: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), volume 1, pp. 2227-2237. - source

      Not checked in the free run.

    • cites [4] [4] J. Howard, S. Ruder, Universal language model finetuning for text classification, in: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pp. 328-339. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [5] [5] A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, Improving language understanding by generative pre-training (2018). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [6] [6] W. Xiong, L. Wu, F. Alleva, J. Droppo, X. Huang, A. Stolcke, The Microsoft 2017 Conversational speech recognition system, in: Proc. Speech and Signal Processing (ICASSP) 2018 IEEE Int. Conf. Acoustics, pp. 5934-5938. [54 http://p4mi.org](http://p4mi.org/) - source

      Not checked in the free run.

    • cites [7] [7] A. van den Oord, S. Dieleman, H. Zen, K. Simonyan, O. Vinyals, A. Graves, N. Kalchbrenner, A. Senior, K. Kavukcuoglu, WaveNet: A generative model for raw audio, arXiv preprint arXiv:1609.03499v2 (2016). - source

      Not checked in the free run.

    • cites [8] [8] C. Guo, F. Berkhahn, Entity embeddings of categorical variables, arXiv preprint arXiv:1604.06737 (2016). - source

      Not checked in the free run.

    • cites [9] [9] A. De Br´ ebisson, ´ E. Simon, A. Auvolat, P. Vincent, Y. Bengio, Artificial neural networks applied to taxi destination prediction, arXiv preprint arXiv:1508.00021 (2015). - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given
    § 2.2. Deep learning · p. 5

    In deep learning, the computers learn useful representations and features automatically, directly from the raw data, bypassing manual feature engineering.

    “In deep learning, the computers learn useful representations and features automatically, directly from the raw data, bypassing this manual and difficult step.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 6

    CNNs were applied in medical image analysis already in the early 90s, but with limited success.

    “Interestingly, CNNs was applied in medical image analysis already in the early 90s, e.g. [58], but with limited success.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [58] [58] S. C. Lo, M. T. Freedman, J. S. Lin, S. K. Mun, Automatic lung nodule detection using profile matching and back-propagation neural network techniques., Journal of digital imaging 6 (1993) 4854. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 1. Introduction · p. 2

    Healthcare applications of deep learning range from one-dimensional biosignal analysis and prediction of medical events to computer-aided detection and diagnosis, drug discovery, therapy selection, operational efficiency, stratified care delivery, and analysis of electronic health records.

    “Healthcare applications of deep learning range from one-dimensional biosignal analysis [17] and the prediction of medical events, e.g. seizures [18] and cardiac arrests [19], to computer-aided detection [20] and diagnosis [21] supporting clinical decision making and survival analysis [22], to drug discovery [23] and as an aid in therapy selection and pharmacogenomics [24], to increased operational”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [17] [17] N. Ganapathy, R. Swaminathan, T. M. Deserno, Deep learning on 1-D biosignals: a taxonomy-based survey, Yearbook of medical informatics 27 (2018) 98-109. - source

      Not checked in the free run.

    • cites [20] [20] H.-C. Shin, H. R. Roth, M. Gao, L. Lu, Z. Xu, I. Nogues, J. Yao, D. Mollura, R. M. Summers, Deep convolutional neural networks for computeraided detection: Cnn architectures, dataset characteristics and transfer learning., IEEE transactions on medical imaging 35 (2016) 1285-1298. - source

      Not checked in the free run.

    • cites [23] [23] J. Jim´ enez, M. ˇ Skaliˇ c, G. Mart´ ınez-Rosell, G. De Fabritiis, KDEEP: Protein-Ligand absolute binding affinity prediction via 3D-Convolutional Neural Networks, Journal of Chemical Information and Modeling 58 (2018) 287-296. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.2. Deep learning · p. 6

    In medical imaging the interest in deep learning is mostly triggered by convolutional neural networks (CNNs), a powerful way to learn useful representations of images and other structured data.

    “In medical imaging the interest in deep learning is mostly triggered by convolutional neural networks (CNNs) [57] 15 , a powerful way to learn useful representations of images and other structured data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [57] [57] Y. LeCun, L. Bottou, Y. Bengio, P. Haffner, Gradient-based learning applied to document recognition, Proceedings of the IEEE 86 (1998) 22782324. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3. Deep learning, medical imaging and MRI · p. 11

    Deep learning is applied in PET-MRI attenuation correction.

    “Deep learning is applied in radiotherapy [89], in PET-MRI attenuation correction [90, 91], in radiomics [92, 93]”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [90] [90] F. Liu, H. Jang, R. Kijowski, T. Bradshaw, A. B. McMillan, Deep learning MR imaging-based attenuation correction for PET/MR imaging, Radiology 286 (2018) 676-684. - source

      Not checked in the free run.

    • cites [91] [91] A. Mehranian, H. Arabi, H. Zaidi, Vision 20/20: Magnetic resonance imaging-guided attenuation correction in pet/mri: Challenges, solutions, and opportunities., Medical physics 43 (2016) 11301155. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 1. Introduction · p. 2

    There are several strong initiatives across academia, and multiple large companies are pursuing healthcare projects based on machine learning.

    “there are several strong initiatives across academia, and multiple large companies are pursuing healthcare projects based on machine learning.”
    Quote in the paperNot a substantial claim0 of 6 sources checked
    • cites [29] [29] V. Gulshan, L. Peng, M. Coram, M. C. Stumpe, D. Wu, A. Narayanaswamy, S. Venugopalan, K. Widner, T. Madams, J. Cuadros, et al., Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs, Jama 316 (2016) 2402-2410. - source

      Not checked in the free run.

    • cites [30] [30] R. Poplin, A. V. Varadarajan, K. Blumer, Y. Liu, M. McConnell, G. Corrado, L. Peng, D. Webster, Predicting Cardiovascular Risk Factors in Retinal Fundus Photographs using Deep Learning, Nature Biomedical Engineering (2018). - source

      Not checked in the free run.

    • cites [31] [31] R. Poplin, P.-C. Chang, D. Alexander, S. Schwartz, T. Colthurst, A. Ku, D. Newburger, J. Dijamco, N. Nguyen, P. T. Afshar, S. S. Gross, L. Dorfman, C. Y. McLean, M. A. DePristo, A universal SNP and small-indel variant caller using deep neural networks, Nature Biotechnology (2018). - source

      Not checked in the free run.

    • cites [32] [32] J. De Fauw, J. R. Ledsam, B. Romera-Paredes, S. Nikolov, N. Tomasev, S. Blackwell, H. Askham, X. Glorot, B. O'Donoghue, D. Visentin, et al., Clinically applicable deep learning for diagnosis and referral in retinal disease, Nature medicine 24 (2018) 1342. - source

      Not checked in the free run.

    • cites [33] [33] Y. Qin, K. Kamnitsas, S. Ancha, J. Nanavati, G. Cottrell, A. Criminisi, A. Nori, Autofocus Layer for Semantic Segmentation, arXiv preprint arXiv:1805.08403 (2018). - source

      Not checked in the free run.

    • cites [34] [34] K. Kamnitsas, C. Baumgartner, C. Ledig, V. Newcombe, J. Simpson, A. Kane, D. Menon, A. Nori, A. Criminisi, D. Rueckert, et al., Unsupervised domain adaptation in brain lesion segmentation with adversarial networks, in: International Conference on Information Processing in Medical Imaging, Springer, pp. 597-609. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • Rests on an assumption: The use of machine learning in general and deep learning in particular within healthcare is still in its infancy.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 7

    Removing the pooling layers simplifies the network architecture without necessarily sacrificing performance.

    “Removing the pooling layers simplifies the network architecture without necessarily sacrificing performance [62].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [62] [62] J. T. Springenberg, A. Dosovitskiy, T. Brox, M. Riedmiller, Striving for simplicity: The all convolutional net, arXiv preprint arXiv:1412.6806 (2014). -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 9

    NASNet was a CNN architecture designed by a neural network, beating all the previous human-designed networks at the ILSVRC competition.

    “A CNN architecture designed by a neural network, beating all the previous human-designed networks at the ILSVRC competition.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [77] [77] B. Zoph, V. Vasudevan, J. Shlens, Q. V. Le, Learning transferable architectures for scalable image recognition, arXiv preprint arXiv:1707.07012 2 (2017). -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 7

    This makes it possible for the entire neural network to approximate almost any nonlinear function.

    “This makes it possible for the entire neural network to approximate almost any nonlinear function [48, 49] 19”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [48] [48] M. Leshno, V. Y. Lin, A. Pinkus, S. Schocken, Multilayer feedforward networks with a nonpolynomial activation function can approximate any function, Neural networks 6 (1993) 861-867. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [49] [49] S. Sonoda, N. Murata, Neural network with unbounded activation functions is universal approximator, Applied and Computational Harmonic Analysis 43 (2017) 233-268. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 8

    In the early days of modern deep learning one tended to use very simple combinations of the building blocks, as in Lenet and AlexNet.

    “In the early days of modern deep learning one tended to use very simple combinations of the building blocks, as in Lenet [57] and AlexNet [1].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [57] [57] Y. LeCun, L. Bottou, Y. Bengio, P. Haffner, Gradient-based learning applied to document recognition, Proceedings of the IEEE 86 (1998) 22782324. - source

      Not checked in the free run.

    • cites [1] [1] A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, in: F. Pereira, C. J. C. Burges, L. Bottou, K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 25, Curran Associates, Inc., 2012, pp. 1097-1105. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 9

    A 152 layer deep ResNet won the 2015 ILSVRC competition, and the authors also successfully trained a version with 1001 layers.

    “A 152 layer deep ResNet won the 2015 ILSVRC competition, and the authors also successfully trained a version with 1001 layers.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [72] [72] K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770-778. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3. Deep learning, medical imaging and MRI · p. 11

    Deep learning is applied in radiomics.

    “Deep learning is applied in radiotherapy [89], in PET-MRI attenuation correction [90, 91], in radiomics [92, 93]”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [92] [92] J. Lao, Y. Chen, Z.-C. Li, Q. Li, J. Zhang, J. Liu, G. Zhai, A deep learning-based radiomics model for prediction of survival in glioblastoma multiforme., Scientific reports 7 (2017) 10353. - source

      Not checked in the free run.

    • cites [93] [93] L. Oakden-Rayner, G. Carneiro, T. Bessen, J. C. Nascimento, A. P. Bradley, L. J. Palmer, Precision radiology: Predicting longevity using feature engineering and deep learning methods in a radiomics framework., Scientific reports 7 (2017) 1648. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 8

    AlexNet launched the current deep learning boom by winning the 2012 ILSVRC competition by a huge margin.

    “The network that launched the current deep learning boom by winning the 2012 ILSVRC competition by a huge margin.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [1] [1] A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, in: F. Pereira, C. J. C. Burges, L. Bottou, K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 25, Curran Associates, Inc., 2012, pp. 1097-1105. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning
    § 1. Introduction · p. 2

    Deep learning techniques have become the de facto standard for a wide variety of computer vision problems.

    “Deep learning techniques have become the de facto standard for a wide variety of computer vision problems.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisereasoning
    § 2.1. Artificial neural networks · p. 4

    By now, artificial neural networks form one of the dominant methods in machine learning, and the most intensively studied.

    “But by now, artificial neural networks form one of the dominant methods in machine learning, and the most intensively studied.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given
    § 2.2. Deep learning · p. 6

    Once it became possible to use features learned directly from the data, many of the handcrafted image features were typically left by the wayside as they turned out to be almost worthless compared to feature detectors found by CNNs.

    “Once it became possible to use features learned directly from the data, many of the handcrafted image features were typically left by the wayside as they turned out to be almost worthless compared to feature detectors found by CNNs.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given
    § 3. Deep learning, medical imaging and MRI · p. 10

    Deep learning methods are increasingly used to improve clinical practice.

    “Deep learning methods are increasingly used to improve clinical practice, and the list of examples is long, growing daily.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 8

    Including batch normalization layers forces the network to periodically change its activations to zero mean and unit standard deviation as the training batch hits these layers, which works as a regularizer for the network, speeds up training, and makes it less dependent on careful parameter initialization.

    “Including batch normalization layers forces the network to periodically change its activations to zero mean and unit standard deviation as the training batch hits these layers, which works as a regularizer for the network, speeds up training, and makes it less dependent on careful parameter initialization [67].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [67] [67] S. Ioffe, C. Szegedy, Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, in: International Conference on Machine Learning, pp. 448-456. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 2.3. Building blocks of convolutional neural networks · p. 9

    ResNet introduced skip connections, which makes it possible to train much deeper networks.

    “Introduced skip connections , which makes it possible to train much deeper networks.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [72] [72] K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770-778. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.1.5. Image synthesis · p. 16

    GANs can be used for biological image synthesis and text-to-image synthesis.

    “These can also be used for biological image synthesis [208, 209] and text-to-image synthesis [210, 211, 212].”
    Quote in the paperNot a substantial claim0 of 5 sources checked
    • cites [208] [208] A. Osokin, A. Chessel, R. E. C. Salas, F. Vaggi, Gans for biological image synthesis, in: Proc. IEEE Int. Conf. Computer Vision (ICCV), pp. 2252-2261. -

      Not checked in the free run.

    • cites [209] [209] G. Antipov, M. Baccouche, J. Dugelay, Face aging with conditional generative adversarial networks, in: Proc. IEEE Int. Conf. Image Processing (ICIP), pp. 2089-2093. - source

      Not checked in the free run.

    • cites [210] [210] C. Bodnar, Text to image synthesis using generative adversarial networks, arXiv preprent arXiv:1805.00676v1 (2018). - source

      Not checked in the free run.

    • cites [211] [211] H. Dong, S. Yu, C. Wu, Y. Guo, Semantic image synthesis via adversarial learning, arXiv preprint arXiv:1707.06873v1 (2017). -

      Not checked in the free run.

    • cites [212] [212] S. Reed, Z. Akata, X. Yan, L. Logeswaran, B. Schiele, H. Lee, Generative adversarial text to image synthesis, arXiv preprint arXiv:1605.05396v2 (2016). -

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 3.1.1. Data acquisition and image reconstruction · p. 12

    Convolutional recurrent neural networks for dynamic MR image reconstruction reconstruct good quality cardiac MR images from highly undersampled complex-valued k-space data by learning spatio-temporal dependencies, outperforming 3D CNN approaches and compressed sensing-based dynamic MRI reconstruction algorithms in computational complexity, reconstruction accuracy and speed for different undersam-

    “convolutional recurrent neural networks for dynamic MR image reconstruction [135], reconstructing good quality cardiac MR images from highly undersampled complex-valued k -space data by learning spatio-temporal dependencies, outperforming 3D CNN approaches and compressed sensing-based dynamic MRI reconstruction algorithms in computational complexity, reconstruction accuracy and speed for different”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [135] [135] C. Qin, J. V. Hajnal, D. Rueckert, J. Schlemper, J. Caballero, A. N. Price, Convolutional recurrent neural networks for dynamic mr image reconstruction., IEEE transactions on medical imaging (2018). - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.1. Data acquisition and image reconstruction · p. 12

    AUTOMAP formulates image reconstruction generically as a data-driven supervised learning task that generates a mapping between the sensor and the image domain based on an appropriate collection of training data.

    “A unified framework for image reconstruction [140], called automated transform by manifold approximation (AUTOMAP) consisting of a feedforward deep neural network with fully connected layers followed by a sparse convolutional autoencoder, formulate image reconstruction generically as a data-driven supervised learning task that generates a mapping between the sensor and the image domain based on an”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [140] [140] B. Zhu, J. Z. Liu, S. F. Cauley, B. R. Rosen, M. S. Rosen, Image reconstruction by domain-transform manifold learning., Nature 555 (2018) 487-492. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.5. Image synthesis · p. 16

    The last couple of years have seen impressive results for photo-realistic image synthesis using deep learning techniques, especially generative adversarial networks.

    “The last couple of years have seen impressive results for photo-realistic image synthesis using deep learning techniques, especially generative adversarial networks (GANs, introduced by Goodfellow et al. in 2014 [80]), e.g. [205, 206, 207].”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [80] [80] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y. Bengio, Generative Adversarial Nets, in: Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems 27, Curran Associates, Inc., 2014, pp. 2672-2680. -

      Not checked in the free run.

    • cites [205] [205] A. Creswell, T. White, V. Dumoulin, K. Arulkumaran, B. Sengupta, A. A. Bharath, Generative adversarial networks: An overview, IEEE Signal Processing Magazine 35 (2018) 53-65. -

      Not checked in the free run.

    • cites [206] [206] Y. Hong, U. Hwang, J. Yoo, S. Yoon, How generative adversarial networks and their variants work: An overview of GAN, arXiv preprint arXiv:1711.05914v7 (2017). -

      Not checked in the free run.

    • cites [207] [207] H. Huang, P. S. Yu, C. Wang, An introduction to image synthesis with generative adversarial nets, arXiv preprint arXiv:1803.04469v1 (2018). -

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 3.1.4. Image super-resolution · p. 16

    Super-resolution for MRI can be used to improve the trade-off between resolution, SNR, and acquisition time, generate 7T-like MR images on 3T MRI scanners, or obtain super-resolution T1 maps from a set of low resolution T1 weighted images.

    “Super-resolution for MRI have been around for almost 10 years [192, 193] and can be used to improve the trade-off between resolution, SNR, and acquisition time [194], generate 7T-like MR images on 3T MRI scanners [195], or obtain super-resolution T 1 maps from a set of low resolution T 1 weighted images [196].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [194] [194] E. Plenge, D. H. J. Poot, M. Bernsen, G. Kotek, G. Houston, P. Wielopolski, L. van der Weerd, W. J. Niessen, E. Meijering, Super-resolution methods in mri: can they improve the trade-off between resolution, signal-to-noise ratio, and acquisition time?, Magnetic resonance in medicine 68 (2012) 19831993. - source

      Not checked in the free run.

    • cites [195] [195] K. Bahrami, F. Shi, I. Rekik, Y. Gao, D. Shen, 7Tguided super-resolution of 3T MRI, Medical physics 44 (2017) 1661-1677. - source

      Not checked in the free run.

    • cites [196] [196] G. Van Steenkiste, D. H. J. Poot, B. Jeurissen, A. J. den Dekker, F. Vanhevel, P. M. Parizel, J. Sijbers, Super-resolution t, javax.xml.bind.jaxbelement@1458c115, estimation: Quantitative high resolution t, javax.xml.bind.jaxbelement@62365375, mapping from a set of low resolution t, javax.xml.bind.jaxbelement@20656587, -weighted images with different slice orientations., Magnetic resonance in medicine 77 (2017) 1818-1830. - source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given
    § 3.1.5. Image synthesis · p. 16

    Generative models for image synthesis serve as a form of data augmentation, and also as an anonymization tool.

    “Such generative models for image synthesis serve as a form of data augmentation, and also as an anonymization tool.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.1. Data acquisition and image reconstruction · p. 12

    Research on CNN and RNN-based image reconstruction methods is rapidly increasing.

    “Research on CNN and RNN-based image reconstruction methods is rapidly increasing, pioneered by Yang et al. [133] at NIPS 2016 and Wang et al. [134] at ISBI 2016.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [133] [133] Y. Yang, J. Sun, H. Li, Z. Xu, Deep ADMMNet for compressive sensing MRI, in: D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon, R. Garnett (Eds.), Advances in Neural Information Processing Systems 29, Curran Associates, Inc., 2016, pp. 1018. -

      Not checked in the free run.

    • cites [134] [134] S. Wang, Z. Su, L. Ying, X. Peng, S. Zhu, F. Liang, D. Feng, D. Liang, Accelerating magnetic resonance imaging via deep learning, in: Biomedical Imaging (ISBI), 2016 IEEE 13th International Symposium on, IEEE, pp. 514-517. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3. Deep learning, medical imaging and MRI · p. 11

    Deep learning is applied for theranostics in neurosurgical imaging, combining confocal laser endomicroscopy with deep learning models for automatic detection of intraoperative CLE images on-the-fly.

    “and for theranostics in neurosurgical imaging, combining confocal laser endomicroscopy with deep learning models for automatic detection of intraoperative CLE images on-the-fly [95].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [95] [95] M. Izadyyazdanabadi, E. Belykh, M. A. Mooney, J. M. Eschbacher, P. Nakaji, Y. Yang, M. C. Preul, Prospects for theranostics in neurosurgical imaging: Empowering confocal laser endomicroscopy diagnostics via deep learning., Frontiers in oncology 8 (2018) 240. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.2. Quantitative parameters - QSM and MR fingerprinting · p. 13

    Quantitative susceptibility mapping (QSM) aims to noninvasively estimate the magnetic susceptibility of biological tissue.

    “Quantitative susceptibility mapping (QSM) is a growing field of research in MRI, aiming to noninvasively estimate the magnetic susceptibility of biological tissue [145, 146].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [145] [145] A. Deistung, A. Sch¨ afer, F. Schweser, U. Biedermann, R. Turner, J. R. Reichenbach, Toward in vivo histology: a comparison of quantitative susceptibility mapping (QSM) with magnitude-, phase-, and R2*-imaging at ultra-high magnetic field strength, Neuroimage 65 (2013) 299-314. - source

      Not checked in the free run.

    • cites [146] [146] A. Deistung, F. Schweser, J. R. Reichenbach, Overview of quantitative susceptibility mapping, NMR in Biomedicine 30 (2017). - source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 3.1.2. Quantitative parameters - QSM and MR fingerprinting · p. 14

    Hoppe et al. found matching one MRF time series was about 100 times faster than the conventional matching method and with very small mean absolute deviations from ground truth values.

    “Matching one time series was about 100 times faster than the conventional [150] matching method and with very small mean absolute deviations from ground truth values.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [166] [166] E. Hoppe, G. Krzdrfer, T. Wrfl, J. Wetzl, F. Lugauer, J. Pfeuffer, A. Maier, Deep learning for magnetic resonance fingerprinting: A new approach for predicting quantitative parameter values from time series, Studies in health technology and informatics 243 (2017) 202-206. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.1.2. Quantitative parameters - QSM and MR fingerprinting · p. 13

    DeepQSM, a U-Net-based convolutional neural network trained on synthetically generated data, can generalize to real-world clinical brain MRI data, computing susceptibility maps within seconds end-to-end.

    “After training their model on synthetic data it was able to generalize to real-world clinical brain MRI data, computing susceptibility maps within seconds end-to-end.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [149] [149] K. G. B. Rasmussen, M. J. Kristensen, R. G. Blendal, L. R. Ostergaard, M. Plocharski, K. O'Brien, C. Langkammer, A. Janke, M. Barth, S. Bollmann, DeepQSM-Using Deep Learning to Solve the Dipole Inversion for MRI Susceptibility Mapping, Biorxiv (2018) 278036. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.1.1. Data acquisition and image reconstruction · p. 12

    Schlemper et al. created a deep cascade of concatenated CNNs for dynamic MR image reconstruction, making use of data augmentation to increase the variation of examples seen by the network and reduce overfitting.

    “Schlemper et.al. [136] created a deep cascade of concatenated CNNs for dynamic MR image reconstruction, making use of data augmentation, both rigid and elastic deformations, to increase the variation of the examples seen by the network and reduce overfitting 28 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [136] [136] J. Schlemper, J. Caballero, J. V. Hajnal, A. N. Price, D. Rueckert, A deep cascade of convolutional neural networks for dynamic MR image reconstruction., IEEE transactions on medical imaging 37 (2018) 491-503. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § 3.1.5. Image synthesis · p. 16

    Deep learning approaches have accelerated research and the amount of costly training data in image synthesis.

    “In this context, deep learning approaches have accelerated research and the amount of costly training data.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.3. Image restoration (denoising, artifact detection) · p. 15

    Deep learning approaches have been introduced to denoising.

    “Recently, deep learning approaches have been introduced to denoising.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [180] [180] C. Bermudez, A. J. Plassard, T. L. Davis, A. T. Newton, S. M. Resnick, B. A. Landman, Learning implicit brain MRI manifolds with deep learning, Proceedings of SPIE-the International Society for Optical Engineering 10574 (2018). -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [181] [181] A. Benou, R. Veksler, A. Friedman, T. Riklin Raviv, Ensemble of expert deep neural networks for spatiotemporal denoising of contrast-enhanced MRI sequences, Medical image analysis 42 (2017) 145-159. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given
    § 3.1. From image acquisition to image registration · p. 12

    Deep learning in MRI has typically been focused on segmentation and classification of reconstructed magnitude images.

    “Deep learning in MRI has typically been focused on segmentation and classification of reconstructed magnitude images.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.2. Quantitative parameters - QSM and MR fingerprinting · p. 13

    Magnetic resonance fingerprinting (MRF) uses a pseudo-randomized acquisition that causes the signals from different tissues to have a unique signal evolution ('fingerprint') that is a function of the multiple material properties being investigated.

    “It uses a pseudo-randomized acquisition that causes the signals from different tissues to have a unique signal evolution ('fingerprint') that is a function of the multiple material properties being investigated.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [150] [150] D. Ma, V. Gulani, N. Seiberlich, K. Liu, J. L. Sunshine, J. L. Duerk, M. A. Griswold, Magnetic resonance fingerprinting, Nature 495 (2013) 187-192. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1.3. Image restoration (denoising, artifact detection) · p. 15

    Deep learning methods have also been applied to MR artifact detection.

    “Deep learning methods has also been applied to MR artifact detection , e.g. poor quality spectra in MRSI [188]; detection and removal of ghosting artifacts in MR spectroscopy [189]; and automated reference-free detection of patient motion artifacts in MRI [190].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [188] [188] S. S. Gurbani, E. Schreibmann, A. A. Maudsley, J. S. Cordova, B. J. Soher, H. Poptani, G. Verma, P. B. Barker, H. Shim, L. A. D. Cooper, A convolutional neural network to filter artifacts in spectroscopic MRI, Magnetic resonance in medicine 80 (2018) 1765-1775. - source

      Not checked in the free run.

    • cites [189] [189] S. P. Kyathanahally, A. Dring, R. Kreis, Deep learning approaches for detection and removal of ghosting artifacts in MR spectroscopy, Magnetic resonance in medicine 80 (2018) 851-863. - source

      Not checked in the free run.

    • cites [190] [190] T. K¨ ustner, A. Liebgott, L. Mauch, P. Martirosian, F. Bamberg, K. Nikolaou, B. Yang, F. Schick, S. Gatidis, Automated reference-free detection of motion artifacts in magnetic resonance images, MAGMA 31 (2018) 243-256. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.1.1. Data acquisition and image reconstruction · p. 12

    Using variational networks for single-shot fast spin-echo MRI with variable density sampling, Chen et al. enabled real-time (200 ms per section) image reconstruction, outperforming conventional parallel imaging and compressed sensing reconstruction.

    “Using variational networks for single-shot fast spin-echo MRI with variable density sampling, Chen et.al. [137] enabled real-time (200 ms per section) image reconstruction, outperforming conventional parallel imaging and compressed sensing reconstruction.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [137] [137] F. Chen, V. Taviani, I. Malkiel, J. Y. Cheng, J. I. Tamir, J. Shaikh, S. T. Chang, C. J. Hardy, J. M. Pauly, S. S. Vasanawala, Variable-density singleshot fast Spin-Echo MRI with deep learning reconstruction by using variational networks, Radiology (2018) 180445. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given
    § 3.2.1. Image segmentation · p. 18

    Image segmentation has become the biggest target for deep learning approaches in medical imaging.

    “It has an almost 50 years long history, and has become the biggest target for deep learning approaches in medical imaging.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.1.2. Quantitative parameters - QSM and MR fingerprinting · p. 13

    MRF can be regarded as a quantitative multiparametric MRI analysis, and with recent acquisition schemes using a single-shot spiral trajectory with undersampling, whole-brain coverage of T1, T2 and proton density maps can be acquired at 1.2 × 1.2 × 3 mm3 voxel resolution in less than 5 min.

    “It can be regarded as a quantitative multiparametric MRI analysis, and with recent acquisition schemes using a single-shot spiral trajectory with undersampling, whole-brain coverage of T 1 , T 2 and proton density maps can be acquired at 1 . 2 × 1 . 2 × 3 mm 3 voxel resolution in less than 5 min [159].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [159] [159] D. Ma, Y. Jiang, Y. Chen, D. McGivney, B. Mehta, V. Gulani, M. Griswold, Fast 3D magnetic resonance fingerprinting for a whole-brain coverage, Magnetic resonance in medicine 79 (2018) 21902197. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 5.1. Data · p. 26

    It is possible to add attention mechanisms to neural networks, enabling them to focus their resources on the most informative components of each layer input.

    “It is also possible to add attention mechanisms to neural networks [346, 347], enabling them to focus their resources on the most informative components of each layer input.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [346] [346] V. Mnih, N. Heess, A. Graves, et al., Recurrent models of visual attention, in: Advances in neural information processing systems, pp. 2204-2212. -

      Not checked in the free run.

    • cites [347] [347] K. Xu, J. Ba, R. Kiros, K. Cho, A. Courville, R. Salakhudinov, R. Zemel, Y. Bengio, Show, attend and tell: Neural image caption generation with visual attention, in: International conference on machine learning, pp. 2048-2057. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation
    § 5.1. Data · p. 26

    Deep neural networks are very inefficient models, requiring large numbers of training samples before they can produce anything remotely useful.

    “These are very inefficient models, requiring large number of training samples before they can produce anything remotely useful”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 3.3. Content-based image retrieval · p. 22

    Deep learning methods have been used for automated generation of radiology reports, typically incorporating LSTM network models to generate textual paragraphs, and to identify findings in radiology reports.

    “Recently, deep learning methods have also been used for automated generation of radiology reports , typically incorporating long-short-term-memory (LSTM) network models to generate the textual paragraphs [314, 315, 316, 317], and also to identify findings in radiology reports [318, 319, 320].”
    Quote in the paperNot a substantial claim0 of 7 sources checked
    • cites [314] [314] B. Jing, P. Xie, E. Xing, On the automatic generation of medical imaging reports, arXiv preprint arXiv:1711.08195v3 (2017). - source

      Not checked in the free run.

    • cites [315] [315] C. Y. Li, X. Liang, Z. Hu, E. P. Xing, Hybrid retrieval-generation reinforced agent for medical image report generation, arXiv preprint arXiv:1805.08298v1 (2018). - source

      Not checked in the free run.

    • cites [316] [316] M. Moradi, A. Madani, Y. Gur, Y. Guo, T. SyedaMahmood, Bimodal network architectures for automatic generation of image annotation from text, arXiv preprint arXiv:1809.01610v1 (2018). - source

      Not checked in the free run.

    • cites [317] [317] Y. Zhang, D. Y. Ding, T. Qian, C. D. Manning, C. P. Langlotz, Learning to summarize radiology findings, arXiv preprint arXiv:1809.04698v1 (????). - source

      Not checked in the free run.

    • cites [318] [318] E. Pons, L. M. M. Braun, M. G. M. Hunink, J. A. Kors, Natural language processing in radiology: A systematic review, Radiology 279 (2016) 329-343. - source

      Not checked in the free run.

    • cites [319] [319] J. Zech, M. Pain, J. Titano, M. Badgeley, J. Schefflein, A. Su, A. Costa, J. Bederson, J. Lehar, E. K. Oermann, Natural language-based machine learning models for the annotation of clinical radiology reports, Radiology 287 (2018) 570-580. - source

      Not checked in the free run.

    • cites [320] [320] D. J. Goff, T. W. Loehfelm, Automated radiology report summarization using an open-source natural language processing pipeline, Journal of digital imaging 31 (2018) 185-192. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 5.1. Data · p. 25

    There are a few examples of federated learning ideas entering the medical machine learning community, as in a study where the distribution of deep learning models among several medical institutions was investigated without considering privacy issues.

    “There are a few examples of these ideas entering the medical machine learning community, as in [340] where the distribution of deep learning models among several medical institutions was investigated, but then without considering the above privacy issues.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [340] [340] K. Chang, N. Balachandar, C. Lam, D. Yi, J. Brown, A. Beers, B. Rosen, D. L. Rubin, J. Kalpathy-Cramer, Distributed deep learning networks among institutions for medical imaging, Journal of the American Medical Informatics Association : JAMIA 25 (2018) 945-954. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § 5.3. Workflow integration, regulations · p. 27

    Even if there is interest on the ground floor and prototype systems get into the hands of clinicians, there are many higher-ups to convince and regulatory, ethical and legal hurdles to overcome.

    “But still, even if there is interest on the 'ground floor' and one is able to get prototype systems into the hands of clinicians, there are many higher-ups to convince and regulatory, ethical and legal hurdles to overcome.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning
    § 5.1. Data · p. 26

    The networks that are most frequently used, and with the best raw performance, remain the data-hungry standard deep neural networks.

    “However, the networks that are most frequently used, and with the best raw performance, remain the data-hungry standard deep neural networks.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning
    § 5.1. Data · p. 26

    For two-dimensional images, one will almost always use a network that has been pre-trained on the ImageNet data set.

    “For two-dimensional images one will almost always use a network that has been pre-trained on the ImageNet data set.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5. Challenges, limitations and future perspectives · p. 25

    The ideas behind state-of-the-art two-dimensional CNNs can be lifted to three dimensions, but adding a third spatial dimension results in additional constraints.

    “It is clear that the ideas behind stateof-the-art two-dimensional CNNs can be lifted to three dimensions, but also that adding a third spatial dimension results in additional constraints.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 4. Open science and reproducible research in machine learning for medical imaging · p. 22

    What is true about the openness of machine learning in general is increasingly true also for the sub-field of machine learning for medical image analysis.

    “What's true about the openness of machine learning in general is increasingly true also for the sub-field of machine learning for medical image analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given
    § 5.2. Interpretability, trust and safety · p. 27

    The field of Bayesian Deep Learning aims to combine deep learning and Bayesian approaches to uncertainty.

    “The field of Bayesian Deep Learning aims to combine deep learning and Bayesian approaches to uncertainty.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 5.3. Workflow integration, regulations · p. 27

    Attempting to augment already established procedures necessitates knowledge of the entire workflow.

    “Attempting to augment already established procedures necessitates knowledge of the entire workflow.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning
    § 5.3. Workflow integration, regulations · p. 27

    It is possible to develop clever machine learning systems for clinical use that turn out to be practically useless for actual clinicians.

    “It is possible to end up developing clever machine learning system for clinical use that turn out to be practically useless for actual clinicians.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 5.1. Data · p. 26

    Pre-training entire models has recently been demonstrated to greatly impact the performance of natural language processing systems.

    “The idea of pre-training networks is not restricted to images. Pre-training entire models has recently been demonstrated to greatly impact the performance of natural language processing systems [3, 4, 5].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [3] [3] M. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, L. Zettlemoyer, Deep contextualized word representations, in: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), volume 1, pp. 2227-2237. -

      Not checked in the free run.

    • cites [4] [4] J. Howard, S. Ruder, Universal language model finetuning for text classification, in: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pp. 328-339. -

      Not checked in the free run.

    • cites [5] [5] A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, Improving language understanding by generative pre-training (2018). -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § 5.1. Data · p. 25

    Most current work on deep learning for medical data analysis uses either open, anonymized data sets or locally obtained anonymized research data, making privacy issues less relevant.

    “Most current work on deep learning for medical data analysis use either open, anonymized data sets (as those in Table 4), or locally obtained anonymized research data, making these issues less relevant.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 3.3. Content-based image retrieval · p. 21

    Sklan et al. trained a CNN for content-based image retrieval with more than one million random MR and CT images, achieving a true positive rate of 20% on an independent test set of 2100 labeled images.

    “One of the first application of deep learning for CBIR in the medical domain came in 2015 when Sklan et al. [310] trained a CNN to perform CBIR with more than one million random MR and CT images, with disappointing results (true positive rate of 20%) on their independent test set of 2100 labeled images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [310] [310] J. E. S. Sklan, A. J. Plassard, D. Fabbri, B. A. Landman, Toward content based image retrieval with deep convolutional neural networks, Proceedings of SPIE-the International Society for Optical Engineering 9417 (2015). -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning
    § 5. Challenges, limitations and future perspectives · p. 25

    The story for deep learning in medical imaging is not quite as settled as for standard computer vision.

    “However, the story for deep learning in medical imaging is not quite as settled.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given
    § 5.2. Interpretability, trust and safety · p. 27

    Making deep neural networks produce robust uncertainty estimates in addition to predictions can increase their trustworthiness.

    “Another way to increase their trustworthiness is to make them produce robust uncertainty estimates in addition to predictions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 5. Challenges, limitations and future perspectives · p. 25

    Using basic building blocks placed according to the ideas behind ResNet and SENet will easily result in close to state-of-the-art performance on two-dimensional object detection, image classification and segmentation tasks.

    “Using the basic building blocks described above, placed according to the ideas behind, say, ResNet and SENet, will easily result in close to state-of-the-art performance on two-dimensional object detection, image classification and segmentation tasks.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 5.4. Perspectives and future expectations · p. 27

    The tremendous amounts of high-impact publications in top journals dealing with deep learning in medical imaging illustrate the value of the methods.

    “This is illustrated by the tremendous amounts of high-impact publications in top-journals dealing with deep learning in medical imaging (for example [17, 21, 30, 32, 40, 90, 162, 137, 140, 273, 285], all published in 2018).”
    Quote in the paperNot a substantial claim0 of 10 sources checked
    • cites [17] [17] N. Ganapathy, R. Swaminathan, T. M. Deserno, Deep learning on 1-D biosignals: a taxonomy-based survey, Yearbook of medical informatics 27 (2018) 98-109. - source

      Not checked in the free run.

    • cites [21] [21] D. S. Kermany, M. Goldbaum, W. Cai, C. C. S. Valentim, H. Liang, S. L. Baxter, A. McKeown, G. Yang, X. Wu, F. Yan, J. Dong, M. K. Prasadha, J. Pei, M. Y. L. Ting, J. Zhu, C. Li, S. Hewett, J. Dong, I. Ziyar, A. Shi, R. Zhang, L. Zheng, R. Hou, W. Shi, X. Fu, Y. Duan, V. A. N. Huu, C. Wen, E. D. Zhang, C. L. Zhang, O. Li, X. Wang, M. A. Singer, X. Sun, J. Xu, A. Tafreshi, M. A. Lewis, H. Xia, K. Zhang, Identifying medical diagnoses and treatable diseases by image-based deep learning, Cell 172 (2018) 1122-1131.e9. - source

      Not checked in the free run.

    • cites [30] [30] R. Poplin, A. V. Varadarajan, K. Blumer, Y. Liu, M. McConnell, G. Corrado, L. Peng, D. Webster, Predicting Cardiovascular Risk Factors in Retinal Fundus Photographs using Deep Learning, Nature Biomedical Engineering (2018). - source

      Not checked in the free run.

    • cites [32] [32] J. De Fauw, J. R. Ledsam, B. Romera-Paredes, S. Nikolov, N. Tomasev, S. Blackwell, H. Askham, X. Glorot, B. O'Donoghue, D. Visentin, et al., Clinically applicable deep learning for diagnosis and referral in retinal disease, Nature medicine 24 (2018) 1342. - source

      Not checked in the free run.

    • cites [40] [40] G. Hinton, Deep Learning A Technology With the Potential to Transform Health Care (2018) 1-2. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [90] [90] F. Liu, H. Jang, R. Kijowski, T. Bradshaw, A. B. McMillan, Deep learning MR imaging-based attenuation correction for PET/MR imaging, Radiology 286 (2018) 676-684. - source

      Not checked in the free run.

    • cites [162] [162] B. Rieger, M. Akakaya, J. C. Pariente, S. Llufriu, E. Martinez-Heras, S. Weingrtner, L. R. Schad, Time efficient whole-brain coverage with MR fingerprinting using slice-interleaved echo-planarimaging, Scientific reports 8 (2018) 6667. - source

      Not checked in the free run.

    • cites [137] [137] F. Chen, V. Taviani, I. Malkiel, J. Y. Cheng, J. I. Tamir, J. Shaikh, S. T. Chang, C. J. Hardy, J. M. Pauly, S. S. Vasanawala, Variable-density singleshot fast Spin-Echo MRI with deep learning reconstruction by using variational networks, Radiology (2018) 180445. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [140] [140] B. Zhu, J. Z. Liu, S. F. Cauley, B. R. Rosen, M. S. Rosen, Image reconstruction by domain-transform manifold learning., Nature 555 (2018) 487-492. - source

      Not checked in the free run.

    • cites [273] [273] J. Wasserthal, P. Neher, K. H. Maier-Hein, TractSeg - fast and accurate white matter tract segmentation, Neuroimage 183 (2018) 239-253. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning
    § 5.1. Data · p. 25

    Because privacy and data protection is often a requirement when dealing with medical data, new techniques for training models without exposing the underlying training data to the user of the model are necessary.

    “As privacy and data protection is often a requirement when dealing with medical data, new techniques for training models without exposing the underlying training data to the user of the model are necessary.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Privacy and data protection is often a requirement when dealing with medical data.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationnone given
    § 5. Challenges, limitations and future perspectives · p. 25

    Important challenges for deep learning in medical imaging are related to data, trust, interpretability, workflow integration, and regulations.

    “Other important challenges are related to data, trust, interpretability, workflow integration, and regulations, as discussed below.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 5.2. Interpretability, trust and safety · p. 27

    Deep neural networks rely on complicated interconnected hierarchical representations of the training data to produce predictions, making interpretation very difficult; this is the black box problem.

    “As deep neural networks relies on complicated interconnected hierarchical representations of the training data to produce its predictions, interpreting these predictions becomes very difficult. This is the 'black box' problem of deep neural networks [348].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [348] [348] D. Castelvecchi, Can we open the black box of AI?, Nature News 538 (2016) 20. - source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning
    § 5.4. Perspectives and future expectations · p. 27

    Even though there are many challenges associated with introducing deep learning in clinical settings, the methods produce results that are too valuable to discard.

    “Even though there are many challenges associated to the introduction of deep learning in clinical settings, the methods produce results that are too valuable to discard.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 5.2. Interpretability, trust and safety · p. 27

    New ways of computing uncertainty estimates from state-of-the-art deep learning models have been developed, and the field has recently seen renewed interest.

    “The ideas date back to the early 90s [354, 355, 356], but the field has recently seen renewed interest from the machine learning community at large, as new ways of computing uncertainty estimates from state of the art deep learning models have been developed [59, 65, 357].”
    Quote in the paperNot a substantial claim0 of 6 sources checked
    • cites [354] [354] R. M. Neal, Bayesian learning for neural networks, Ph.D. thesis, University of Toronto, 1995. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [355] [355] D. J. MacKay, A practical Bayesian framework for backpropagation networks, Neural computation 4 (1992) 448-472. - source

      Not checked in the free run.

    • cites [356] [356] P. Dayan, G. E. Hinton, R. M. Neal, R. S. Zemel, The Helmholtz machine, Neural computation 7 (1995) 889-904. - source

      Not checked in the free run.

    • cites [59] [59] S. Murray, An exploratory analysis of multi-class uncertainty approximation in Bayesian convolution neural networks, Master's thesis, University of Bergen, 2018. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [65] [65] Y. Gal, Uncertainty in deep learning, Ph.D. thesis, University of Cambridge, 2016. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [357] [357] Y. Li, Y. Gal, Dropout Inference in Bayesian Neural Networks with Alpha-divergences, in: International Conference on Machine Learning, pp. 2052-2061. -

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 5.1. Data · p. 25

    Data access, privacy issues, and data protection are crucially important obstacles for deep neural networks, especially in medical data analysis.

    “This is a crucially important obstacle for deep neural networks, especially in medical data analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given
    § 4. Open science and reproducible research in machine learning for medical imaging · p. 22

    Deep learning-based models are completely dominating the leaderboards of essentially all image-based competitions, and other machine learning or non-machine learning techniques have largely been outclassed.

    “It is interesting to note how deep learning-based models are completely dominating the leaderboards of essentially all image-based competitions. Other machine learning models, or non-machine learning-based techniques, have largely been outclassed.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 5.1. Data · p. 26

    Data augmentation by applying label-preserving transformations is a widely used technique.

    “Another widely used technique is augmenting the training data set by applying various transformations that preserves the labels, as in rotations, scalings and intensity shifts of images, or more advanced data augmentation techniques like anatomically sound deformations, or other data set specific operations (for example in our work on kidney segmentation from DCE-MRI, where we used image”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [343] [343] A. Lundervold, K. Sprawka, A. Lundervold, Fast estimation of kidney volumes and time courses in DCE-MRI using convolutional neural networks, 2018. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given
    § 5.1. Data · p. 26

    The way current deep neural networks are constructed and trained results in significant data size requirements.

    “But whatever you do, the way current deep neural networks are constructed and trained results in significant data size requirements.”
    Quote in the paperNot a substantial claim
73 of 73

  • The paper relies on this without support.

    Healthcare providers generate and capture enormous amounts of data containing extremely valuable signals and information, at a pace far surpassing what 'traditional' methods of analysis can process.

    “Healthcare providers generate and capture enormous amounts of data containing extremely valuable signals and information, at a pace far surpassing what 'traditional' methods of analysis can process.”

  • The paper relies on this without support.

    The use of machine learning in general and deep learning in particular within healthcare is still in its infancy.

    “The use of machine learning in general and deep learning in particular within healthcare is still in its infancy”

  • The paper relies on this without support.

    Medical imaging datasets are often imbalanced, with few pathological findings, limiting the training of deep learning models.

    “This approach is highly interesting as medical imaging datasets are often imbalanced, with few pathological findings, limiting the training of deep learning models.”

  • The paper relies on this without support.

    Medical images are often three-dimensional.

    “One issue is that medical images are often three-dimensional, and three-dimensional convolutional neural networks are as well-developed as their 2D counterparts.”

  • The paper relies on this without support.

    Three-dimensional convolutional neural networks are as well-developed as their 2D counterparts.

    “One issue is that medical images are often three-dimensional, and three-dimensional convolutional neural networks are as well-developed as their 2D counterparts.”

  • The paper relies on this without support.

    Labeled training data is typically both expensive and difficult to produce.

    “These are very inefficient models, requiring large number of training samples before they can produce anything remotely useful, and labeled training data is typically both expensive and difficult to produce.”

  • The paper relies on this without support.

    Privacy and data protection is often a requirement when dealing with medical data.

    “As privacy and data protection is often a requirement when dealing with medical data, new techniques for training models without exposing the underlying training data to the user of the model are necessary.”

  • The paper relies on this without support.

    The training data has to be representative of the data the network will meet in the future.

    “In addition, the training data has to be representative of the data the network will meet in the future.”

  • The paper relies on this without support.

    The high-quality images one typically works with in research differ from the messiness of the real clinical world.

    “Considering the large difference between the high-quality images one typically work with when doing research and the messiness of the real, clinical world, this can be a major obstacle when putting deep learning systems into production.”
9 of 9

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • CNNs can classify images in the right category after learning from existing data, similar to how a radiologist operates.

    “After learning a new capability from the existing data, this can be applied to new images and the system can classify the images in the right category (Inference), similar to how a radiologist operates [28].”

    From The rise of artificial intelligence in healthcare applications (2020) · cites this paper as [28]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Spinal-metastasis entry uses a prostate-cancer reference

    Minor

    Table 2 assigns reference [291] to spinal metastasis detection, whereas reference [291] is a study of prostate-cancer MRI classification.

  • Spondylitis entry points to an image-resolution paper

    Minor

    Table 2 describes [297] as a study distinguishing types of spondylitis, but the listed reference concerns improving MR image resolution.

  • ILSVRC is described as both ended and ongoing

    Minor

    The introduction says the last ILSVRC competition was held in 2017, but Section 4 calls it a "now annual" competition.

  • PET attenuation-correction citation points to Alzheimer's diagnosis

    Minor

    The image-registration section cites [244] for MRI-based PET attenuation correction, but reference [244] is titled as an Alzheimer's disease diagnosis study.

4 of 4

Cites

Flagged first, then by citations

Research works printed but not linked: 28

  1. Shickel, 2017, Deep EHR: a survey of recent advances in deep learning techniques for electronic health record (EHR) analysis, IEEE J Biomed Health Inform
  2. Poplin, 2018, Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning, Nat Biomed Eng, 2, 158, 10.1038/s41551-018-0195-0 · DOI · Paper page
  3. Cauchy, 1847, Méthode générale pour la résolution des systemes déquations simultanées, C R Sci Paris, 25, 536
  4. Srivastava, 2014, Dropout: a simple way to prevent neural networks from overfitting, J Mach Learn Res, 15, 1929
  5. Zheng, 2018, Pairwise domain adaptation module for CNN-based 2-D/3-D registration, J Med Imaging (Bellingham, Wash.), 5, 021204
  6. Savadjiev, 2018, Demystification of AI-driven medical image interpretation: past, present and future, Eur Radiol
  7. Qin, 2018, Convolutional recurrent neural networks for dynamic MR image reconstruction, IEEE Trans Med Imaging
  8. Chen, 2018, Variable-density single-shot fast Spin-Echo MRI with deep learning reconstruction by using variational networks, Radiology, 180445
  9. Knoll, 2018, Assessment of the generalization of learned image reconstruction and the potential for transfer learning, Magn Reson Med
  10. Mardani, 2018, Deep generative adversarial neural networks for compressive sensing (GANCS) MRI, IEEE Trans Med Imaging
  11. Rasmussen, 2018, DeepQSM-using deep learning to solve the dipole inversion for MRI susceptibility mapping, Biorxiv, 278036
  12. Aggarwal, 2018, Modl: model based deep learning architecture for inverse problems, IEEE Trans Med Imaging
  13. Li H, Schwab J, Antholzer S, Haltmeier M. Nett: solving inverse problems with deep neural networks, arXiv:1803.00092 (2018). · arXiv
  14. Hoppe, 2017, Deep learning for magnetic resonance fingerprinting: a new approach for predicting quantitative parameter values from time series, Stud Health Technol Inform, 243, 202
  15. Vincent, 2010, Stacked denoising autoencoders: learning useful representations in a deep network with a local denoising criterion, J Mach Learn Res (JMLR), 11, 3371
  16. Hong Y, Hwang U, Yoo J, Yoon S. How generative adversarial networks and their variants work: an overview of GAN, arXiv:1711.05914v7 (2017). · arXiv
  17. Nie, 2017, Medical image synthesis with context-aware generative adversarial networks, Medical image computing and computer-assisted intervention: MICCAI, 417
  18. Spuhler, 2018, Synthesis of patient-specific transmission image for PET attenuation correction for PET/MR imaging of the brain using a convolutional neural network, J Nucl Med
  19. Torrado-Carvajal, 2018, Dixon-VIBE deep learning (DIVIDE) pseudo-CT synthesis for pelvis PET/MR attenuation correction, J Nucl Med
  20. Bernal J, Kushibar K, Asfaw DS, Valverde S, Oliver A, Mart R, et al. Deep convolutional neural networks for brain image analysis on magnetic resonance imaging: a review, arXiv:1712.03747v3 (2017). · arXiv
  21. Bernal, 2018, Deep convolutional neural networks for brain image analysis on magnetic resonance imaging: a review, Artif Intell Med
  22. Pizarro, 2018, Using deep learning algorithms to automatically identify the brain MRI contrast: implications for managing large databases, Neuroinformatics
  23. Hoseini, 2018, Adaptahead optimization algorithm for learning deep CNN applied to MRI segmentation, J Digit Imaging
  24. Shehata, 2018, Computer-aided diagnostic system for early detection of acute renal transplant rejection using diffusion-weighted MRI, IEEE Trans Bio-med Eng
  25. Sklan, 2015, Toward content based image retrieval with deep convolutional neural networks, Proc SPIE, 9417
  26. Ronneberger, 2015, U-net: Convolutional networks for biomedical image segmentation, vol. 9351, 234
  27. Montavon, 2017, Methods for interpreting and understanding deep neural networks, Digit Signal Process
  28. Hohman, 2018, Visual analytics in deep learning: an interrogative survey for the next frontiers, IEEE Trans Vis Comput Graph

Cited by

Sorted by citations

Authors
Lundervold, A. S., & Lundervold, A.
Affiliations
Western Norway University of Applied Sciences (NO)
Corresponding author
Lundervold, Alexander Selvikvåg
Venue
Zeitschrift für Medizinische Physik
Publisher
Elsevier BV
Published
2019-05
Volume
29
Issue
2
Pages
102-127
ISSN
0939-3889 (ISSN-L)
DOI
10.1016/j.zemedi.2018.11.002
OpenAlex
W2900954917
Semantic Scholar
0ebc300c16f01a4e94c8551997922fdb67ac1951
Cited by
2,058
References
287
Citations by year
2026: 189 · 2025: 291 · 2024: 283 · 2023: 286 · 2022: 368 · 2021: 399 · 2020: 196 · 2019: 44 · 2018: 1 · 2012: 1
FWCI
19.06
Type
journal article
Language
en
Open access
Licence
Locations
  • Zeitschrift für Medizinische Physik (publishedVersion, cc-by)
  • arXiv (Cornell University) (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University)
Funders
Bergens Forskningsstiftelse
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced MRI Techniques and Applications
Keywords
deep learning, medical image analysis, magnetic resonance imaging, deep neural network, MRI segmentation, disease prediction, medical image retrieval, open-source code
MeSH terms
Deep Learning, Animals, Humans, Image Processing, Computer-Assisted, Magnetic Resonance Imaging

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review