CoChat
Sign Up
Cited by 798Medical Image Segmentation TechniquesArtificial Intelligence in Healthcare and EducationAI in cancer detection

Deep Learning in Medical Ultrasound Analysis: A Review

Open access
Liu, Shengfeng, Wang, Yi, Yang, Xin, Lei, Baiying, Liu, Li, Li, Shawn Xiang, Ni, Dong, Wang, Tianfu·2019·Engineering
DOI 10.1016/j.eng.2018.11.020
Open this paper in CoChat — free accountOpen PDF

Abstract

Ultrasound (US) has become one of the most commonly performed imaging modalities in clinical practice. It is a rapidly evolving technology with certain advantages and with unique challenges that include low imaging quality and high variability. From the perspective of image analysis, it is essential to develop advanced automatic US image analysis methods to assist in US diagnosis and/or to make such assessment more objective and accurate. Deep learning has recently emerged as the leading machine learning tool in various research fields, and especially in general imaging analysis and computer vision. Deep learning also shows huge potential for various automatic US image analysis tasks. This review first briefly introduces several popular deep learning architectures, and then summarizes and thoroughly discusses their applications in various specific tasks in US image analysis, such as classification, detection, and segmentation. Finally, the open challenges and potential trends of the future application of deep learning in medical US image analysis are discussed.

AI snapshot

Written from the paper’s full text.

TL;DR

This review surveys deep learning architectures and their use in medical ultrasound image analysis, focusing on classification, detection, segmentation, and 3D ultrasound. It concludes that deep learning has strong potential but remains limited by small annotated datasets, transfer-learning challenges, and the difficulty of 3D deep learning.

Key findings

  • The review covers supervised, unsupervised, and hybrid deep models—mainly CNNs, RNNs, RBMs/DBNs, and AEs—and summarizes their application to ultrasound classification, detection, segmentation, and 3D ultrasound analysis.
  • Reported example results include breast tumor classification reaching 92.4% accuracy with DPN/S-DPN; focal liver lesion classification at 86.36% with a DBN and 90.41% with DCCA+MKL; thyroid nodule integrated features improving accuracy by 14%; and transferred CNNs for fetal anatomical structures achieving 91.5% vs. 87.9% for directly learned CNNs.
  • Segmentation examples include fetal left ventricle dynamic CNN improving mean Dice from 88.35% to 94.5%, kidney segmentation improving approximately 5% with prior shape information, and aortic valve MSDL showing up to 42.5% improvement over the state of the art.
  • The paper identifies transfer learning and domain-specific model design as key strategies for small medical ultrasound datasets, while noting that 3D ultrasound remains promising but computationally and methodologically challenging.

Methods

  • Narrative review: literature search until 2018 Feb 1 using keywords (“ultrasound” OR “ultrasonography” OR “ultrasonic imaging” AND “convolutional” OR “deep learning”) in PubMed, Google Scholar, and conference proceedings including MICCAI, SPIE, ISBI, and EMBC.
  • Abstracts were screened, resulting in nearly 100 relevant papers; the review organizes findings by deep learning architecture and by task (classification, detection, segmentation) across anatomical structures, with a separate section on 3D ultrasound.

Limitations

  • Stated in the paper: limited public medical ultrasound datasets and small training samples are a bottleneck; 3D deep learning is computationally expensive and prone to overfitting; transfer learning risks negative transfer and heterogeneous feature spaces.
  • Stated in the paper: many current 3D ultrasound methods are actually slice-by-slice 2D approaches and cannot exploit anatomical context in orthogonal directions.
  • Evident from the text: as a narrative review, it does not perform a systematic quality assessment or quantitative meta-analysis; the reported performance numbers are illustrative examples from cited studies rather than independently validated pooled results.

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

Integrity signals

Registry

Crossref · 10.1016/j.eng.2018.11.020

Cites

168 of 168 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 74 cited sources checked · 0 supported · 0 flagged

  • Premisecitation

    A study on prenatal detection of malformations using US images demonstrated that sensitivity ranged from 27.5% to 96% among different medical institutes.

    “For example, a study on the prenatal detection of malformations using US images demonstrated that the sensitivity ranged from 27.5% to 96% among different medical institutes [3].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • Premisecitation

    Medical US has high inter- and intra-observer variability across different institutes and manufacturers' US systems.

    “However, US also presents unique challenges, such as low imaging quality caused by noise and artifacts, high dependence on abundant operator or diagnostician experience, and high inter-and intra-observer variability across different institutes and manufacturers' US systems.”
    • Also stated in: § 1. Introduction · boosted
      “For example, a study on the prenatal detection of malformations using US images demonstrated that the sensitivity ranged from 27.5% to 96% among different medical institutes [3].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] source

      Not checked in the free run.

  • Premisecitation

    Few review articles focus on medical US analysis, aside from one or two papers examining specific tasks such as breast US image segmentation.

    “However, few focus on medical US analysis, aside from one or two papers that examine specific tasks such as breast US image segmentation [55].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [55] source

      Not checked in the free run.

  • Premisecitation

    Several review articles on deep learning in medical image analysis focus on the whole field of medical image analysis or on other single-imaging modalities such as MRI and microscopy.

    “Several review articles have been written to date on the application of deep learning to medical image analysis; these articles focus on either the whole field of medical image analysis [11][12][13][14][15] or other single-imaging modalities such as MRI [53] and microscopy [54].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [11][12][13][14][15]

      Not checked in the free run.

    • cites [53] source

      Not checked in the free run.

    • cites [54] source

      Not checked in the free run.

  • Premisecitation

    Antropova et al.'s method fusing pre-trained CNN features with hand-designed features using conventional CADx methods on three clinical imaging modality datasets demonstrated significant performance improvement.

    “In the work of Antropova et al. [91], a method that fuses the extracted and pooled low-and mid-level features using a pre-trained CNN with hand-designed features using conventional CADx methods was applied to three clinical imaging modality datasets, and demonstrated significant performance improvement.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [91] source

      Not checked in the free run.

  • Classified as probably not a substantial claim. 20
  • May combine two claims. 33

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

    Premisereasoning

    US-based diagnostic techniques have become one of the most important noninvasive methods for the detection, diagnosis, intervention, and treatment of liver cancer.

    “Therefore, US-based diagnostic techniques have become one of the most important noninvasive methods for the detection, diagnosis, intervention, and treatment of liver cancer.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Liu et al. and Shi et al. employed deep polynomial network or stacked DPN on two small US datasets and obtained highest classification accuracies of 92.4%, outperforming unsupervised deep learning algorithms such as stacked AE and DBN.

    “Both Liu et al. [87] and Shi et al. [19] employed a supervised deep learning algorithm called a deep polynomial network (DPN), or its stacked version, namely stacked DPN (S-DPN), on two small US datasets.”
    • Also stated in: § 3.1.1. Tumors or lesions · plain
      “the highest classification accuracies of 92.4% is obtained, outperforming the unsupervised deep learning algorithms, such as stacked AE and DBN.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [87]

      Not checked in the free run.

    • cites [19] source

      Not checked in the free run.

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

    Interpretationnone given

    3D US presents a promising trend in improving US imaging diagnosis in clinical practice.

    “Furthermore, 3D US presents a promising trend in improving US imaging diagnosis in clinical practice, which is discussed in detail as a separate sub-section.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Deep learning can directly learn features from images in an automatic manner.

    “In contrast, deep learning can directly learn features from images in an automatic manner.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Biopsy is still the current golden standard for liver cancer diagnosis and is heavily relied upon by conventional CADx methods.

    “Biopsy is still the current golden standard for liver cancer diagnosis, and is heavily relied upon by conventional CADx methods.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Deep learning architectures used in most current works in medical US analysis can be categorized into three major classes: supervised deep networks, unsupervised deep networks, and hybrid deep networks.

    “Based on the deep learning architectures and techniques in question, such as classification, segmentation, or detection, the deep learning architectures that are used in most of the current works in this field can be categorized into three major classes: ① supervised deep networks or deep discriminative models, ② unsupervised deep networks or deep generative models, and ③ hybrid deep networks.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    A 2D US image presents a thin plane at arbitrary angles in the body, and these planes are difficult to localize and reproduce later for follow-up studies.

    “A 2D US image presents a thin plane at various arbitrary angles in the body.These planes are difficult to localize and reproduce later for follow-up studies [51].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [51] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Biopsy is invasive and uncomfortable and can easily cause other adverse effects.

    “However, biopsy is invasive and uncomfortable, and can easily cause other adverse effects.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    The basic models or architectures applied in current medical US analysis are mainly CNNs, RNNs, RBMs/DBNs, AEs, and variants of these architectures.

    “The basic models or architectures applied in current medical US analysis are mainly CNNs, recurrent neural networks (RNNs), RBMs/DBNs (where DBN refers to deep belief networks), AEs, and variants of these deep learning architectures, as shown in Fig. 3.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Deep learning is a state-of-the-art tool for automatic analysis tasks, and its use can lead to marked improvement in performance.

    “thus demonstrating that deep learning is a state-of-the-art tool for the performance of automatic analysis tasks, and that its use can lead to marked improvement in performance.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [8] source

      Not checked in the free run.

    • cites [9]

      Not checked in the free run.

    • cites [11] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    SDAE, PGBM combined with RBM, and GoogLeNet CNN applied to breast US or shear-wave elastography images for breast cancer diagnosis obtained superior performance compared with human experts.

    “In addition, the stacked denoising auto-encoder (SDAE) [88], a combination of the point-wise gated Boltzmann machine (PGBM) and RBM [89], and the GoogLeNet CNN [90] have been applied to breast US or shear-wave elastography images for breast cancer diagnosis; all of these obtained a superior performance when compared with human experts.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [88] source

      Not checked in the free run.

    • cites [89] source

      Not checked in the free run.

    • cites [90] source

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    Two main strategies for transferring knowledge from other domains to medical US are using a pre-trained network as a feature extractor and fine-tuning a pre-trained network on medical US images or video sequences.

    “This process involves two main strategies: ① using a pre-trained network as a feature extractor (i.e., to learn features from scratch); and ② fine-tuning a pre-trained network on medical US images or video sequences-a method that is widely applied at present in US analysis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Diagnostic and therapeutic decisions often require accurate estimation of organ or tumor volume, but 2D US calculates volume from simple measurements assuming an idealized shape, which may lead to low accuracy, high variability, and operator dependency.

    “Diagnostic (e.g., OB-GYN) and therapeutic (e.g., staging and planning) decisions often require accurate estimation of organ or tumor volume; however, 2D US techniques calculate volume from simple measurements of length, width, and height in two orthogonal views by assuming an idealized (e.g., ellipsoidal) shape.This may lead to low accuracy, high variability, and operator dependency.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisereasoning

    Current medical US analysis still focuses on 2D US image processing, but there is a growing trend in applications of deep learning in 3D medical US analysis.

    “Although current medical US analysis still focuses on twodimensional (2D) US image processing, there is a growing trend in applications of deep learning in three-dimensional (3D) medical US analysis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisenone given

    Transfer learning has been widely adopted and exhibits excellent capacity for improving learning performance without the need for large samples.

    “Another key solution is transfer learning, which has also been widely adopted and which exhibits excellent capacity for improving the performance of learning without the need for large samples.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Recent applications of deep learning in medical US analysis have involved classification, segmentation, detection, registration, biometric measurements, quality assessment, image-guided interventions, and therapy.

    “Recent applications of deep learning in medical US analysis have involved various tasks, such as traditional diagnosis tasks including classification, segmentation, detection, registration, biometric measurements, and quality assessment, as well as emerging tasks including image-guided interventions and therapy [16] (Fig. 1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Until recently, RNNs have not been widely utilized in sequence models because it is difficult to train them to capture long-term dependencies, giving rise to gradient explosion or vanishing problems.

    “However, until recently, RNNs have not been widely utilized in the various study tasks that are referred to as sequence models.This is partly because it is difficult to train the RNN itself to capture long-term dependencies, in which the RNN usually gives rise to gradient explosion or gradient vanishing problems that were discovered in the early 1990s [61].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [61] source

      Not checked in the free run.

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

    Premisecitation

    CNN is one of the most popular deep architectures and has gained great success in image classification, object detection, and target segmentation.

    “CNN is known as one of the most popular deep architectures and has also gained great success in various tasks, such as image classification [48,49], object detection [29,30], and target segmentation [44,50].”
    Quote in the paperNot a substantial claim0 of 6 sources checked
    • cites [48] source

      Not checked in the free run.

    • cites [49] source

      Not checked in the free run.

    • cites [29] source

      Not checked in the free run.

    • cites [30]

      Not checked in the free run.

    • cites [44]

      Not checked in the free run.

    • cites [50]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Both pre-trained feature extraction and fine-tuning strategies achieve excellent performance in several specific tasks.

    “Both strategies achieve excellent performance in several specific tasks [77,78].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [77] source

      Not checked in the free run.

    • cites [78] source

      Not checked in the free run.

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

    Premisecitation

    The current application of deep learning in medical US analysis is a growing trend supported by progress in the field.

    “Furthermore, the current application of deep learning in medical US analysis is a growing trend that is supported by progress in this field [23,52].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [23]

      Not checked in the free run.

    • cites [52] source

      Not checked in the free run.

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

    Premisenone given

    Supervised deep models are widely used for classification, segmentation, and detection of anatomical structures in medical US images, with CNNs and RNNs being the two most popular architectures.

    “At present, supervised deep models are widely used for the classification, segmentation, and detection of anatomical structures in medical US images; for these tasks, CNNs and RNNs are the two most popular architectures.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    RNNs are mainly applied in speech or text-recognition domains and are rarely used in medical image analysis, much less medical US analysis.

    “Thus far, RNNs are mainly applied in speech or text-recognition domains, and are rarely used in medical image analysis, much less medical US analysis.”
    Quote in the paperNot a substantial claim

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

    Premisereasoning

    RNNs have an inherent advantage for modeling sequence data such as medical US video sequences.

    “The RNN has an inherent advantage for modeling sequence data (e.g., medical US video sequences) due to the structural characteristics of the network.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisenone given

    Traditional machine learning methods for classification often used handcrafted features with a multi-way linear classifier, but these methods are susceptible to image distortion.

    “Traditional machine learning methods often utilized various handcrafted features extracted from US images in combination with a multi-way linear classifier (e.g., SVM) in order to achieve a specific classification task.However, these methods are susceptible to image distortion, such as deformation due to the internal or external environments, or to conditions in the imaging process.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    The requirement of large labeled training samples is difficult to meet in current medical US analysis, where expert annotation is expensive and some diseases are scarce in datasets.

    “However, this requirement is difficult to meet in current medical US analysis, where expert annotation is expensive and where some diseases (e.g., lesions or nodules) are scarce in the datasets [68].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [68] source

      Not checked in the free run.

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

    Premisecitation

    Following Jamieson et al.'s success, many similar studies applied deep learning methods to breast tumor diagnosis.

    “Following this success, many similar studies applied deep learning methods to breast tumor diagnosis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Jamieson et al. (2012) used adaptive deconvolutional networks to learn image features from breast tumor US images, and the performance reached the level of conventional CADx schemes that employ human-designed features.

    “As early as 2012, Jamieson et al. [86] performed a preliminary study that referred to the use of deep learning in the task of classifying breast tumors or mass lesions.”
    • Also stated in: § 3.1.1. Tumors or lesions · plain
      “The results showed that the performance reached the level of conventional CADx schemes that employ human-designed features.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [86]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Wu et al. applied a three-layer DBN to time-intensity curves from contrast-enhanced US video sequences to classify malignant and benign focal liver lesions and achieved highest accuracy of 86.36%, outperforming LDA, k-NN, SVM, and BPN.

    “Wu et al. [22] applied a three-layer DBN in timeintensity curves (TICs) extracted from contrast-enhanced US (CEUS) video sequences in order to classify malignant and benign focal liver lesions.They achieved a highest accuracy of 86.36%, thus outperforming conventional machine methods such as linear discriminant analysis (LDA), k-nearest neighbors (k-NN), SVM, and back propagation net (BPN).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] source

      Not checked in the free run.

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

    Premisereasoning

    To overcome the limitations of 2D US, a variety of 3D US scanning, reconstruction, and display techniques have been developed, providing a broad foundation for 3D medical US analysis.

    “To overcome the limitations of 2D US, a variety of 3D US scanning, reconstruction, and display techniques have been developed, which provide a broad foundation for 3D medical US analysis.”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data

    To screen the papers from the search, the abstract of every paper was read in detail, and relevant papers were chosen, resulting in nearly 100 relevant papers.

    “To screen the papers resulting from this search, the abstract of every paper was read in detail; papers that were relevant for this review were then chosen, which finally resulted in nearly 100 relevant papers, as summarized in Fig. 2 and Table S1 in Appendix A.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    The great success of deep learning comes from the fact that a large number of labeled training samples are required to achieve excellent learning performance.

    “The great success of deep learning comes from the fact that a large number of labeled training samples are required in order to achieve excellent learning performance.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    RNNs are generally considered a type of supervised deep network used for a variety of tasks in medical US analysis.

    “In practical terms, an RNN is generally considered to be a type of supervised deep network that is used for a variety of tasks in medical US analysis [21,60].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [21]

      Not checked in the free run.

    • cites [60]

      Not checked in the free run.

  • May combine two claims.

    Premisenone given

    Deep neural networks have advantages due to their direct learning of mid- and high-level abstract features from raw data, and can directly output an individual prediction label for each image.

    “Here, deep neural networks (DNNs) have several obvious advantages due to their direct learning of mid-and high-level abstract features from the raw data (or images).In addition, DNNs can be directly used to output an individual prediction label for each image, in order to classify targets of interest.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Because US images are 2D while anatomical structure is 3D, the examiner must integrate multiple images mentally, and lack of this ability leads to variability and incorrect diagnosis or misdiagnosis.

    “Although US images are 2D, the anatomical structure is 3D; thus, the examiner/diagnostician must possess the ability to integrate multiple images in his or her mind (in an often inefficient and time-consuming process).Lack of this ability will lead to variability and incorrect diagnosis or misdiagnosis.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Deep learning has entered various application areas of medical US analysis.

    “From the examples provided above, it is evident that deep learning has entered various application areas of medical US analysis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Wu et al. implanted the FCN core into an auto-context scheme to bridge severe boundary incompleteness and remarkably improve segmentation accuracy.

    “In addition, Wu et al. [141] implanted the FCN core into an auto-context scheme [142] in order to take advantage of local contextual information, and thus bridge severe boundary incompleteness and remarkably improve the segmentation accuracy.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [141]

      Not checked in the free run.

    • cites [142] source

      Not checked in the free run.

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

    Interpretationreasoning

    Although deep learning methods have constantly updated state-of-the-art performance results across different application aspects in medical US analysis, there is still room for improvement.

    “However, although deep learning methods have constantly updated state-of-the-art performance results across different application aspects in medical US analysis, there is still room for improvement.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Acquisition of the standard plane is a prerequisite step and is crucial for subsequent biometric measurements and diagnosis in fetal US diagnosis.

    “Among the workflow of fetal US diagnosis, acquisition of the standard plane is the prerequisite step and is crucial for subsequent biometric measurements and diagnosis [114].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [114] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Using only the CPU, the aortic valve can be successfully segmented in less than one second with higher accuracy than the original MSL.

    “By only using the central processing unit (CPU), the aortic valve can be successfully segmented in less than one second with higher accuracy than the original MSL.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Developing domain-specific deep learning models for medical imaging can improve task-specific performance with low computation complexity and facilitate technological advantages in CADx in the medical imaging domain.

    “Consequently, developing domain-specific deep learning models for medical imaging can not only improve task-specific performance with a low computation complexity, but also facilitate technological advantages in CADx in the medical imaging domain.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    MSDL demonstrated efficiency and robustness in 3D detection and segmentation of the aortic valve, with significant improvement up to 42.5% over the state of the art.

    “The results on a large 3D transesophageal echocardiogram image dataset demonstrated the efficiency and robustness of the MSDL in the 3D detection and segmentation task of the aortic valve; it showed a significant improvement of up to 42.5% over the state of the art.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The combination of sparse manifold learning and DBN in the rigid detection stage yielded performance as accurate as the state of the art but with lower training and search complexity.

    “Their experiments demonstrated that the combination of sparse manifold learning and DBN in the rigid detection stage yielded a performance as accurate as the state of the art, but with lower training and search complexity.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [124]

      Not checked in the free run.

    • cites [135] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Incorrect fetal abdominal circumference measurement may lead to inaccurate fetal weight estimation and further increase the risk of misdiagnosis.

    “In clinical examination and diagnosis, incorrect fetal AC measurement may lead to inaccurate fetal weight estimation and further increase the risk of misdiagnosis [98].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [98] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Nascimento and Carneiro directly performed non-rigid segmentation through sparse low-dimensional manifold mapping of explicit contours, but with limited generalization capability.

    “Unlike the typical non-rigid segmentation scheme, Nascimento and Carneiro [136] directly performed non-rigid segmentation through the sparse lowdimensional manifold mapping of explicit contours, but with a limited generalization capability.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [136]

      Not checked in the free run.

44 of 44

  • The paper relies on this without support.

    Medical US imaging has low imaging quality caused by noise and artifacts.

    “However, US also presents unique challenges, such as low imaging quality caused by noise and artifacts, high dependence on abundant operator or diagnostician experience, and high inter-and intra-observer variability across different institutes and manufacturers' US systems.”

  • The paper relies on this without support.

    Medical US diagnosis has high dependence on abundant operator or diagnostician experience.

    “However, US also presents unique challenges, such as low imaging quality caused by noise and artifacts, high dependence on abundant operator or diagnostician experience, and high inter-and intra-observer variability across different institutes and manufacturers' US systems.”

  • The paper relies on this without support.

    Manual segmentation methods are time consuming and tedious, and suffer from great individual variability.

    “Furthermore, it is well known that manual segmentation methods are time consuming and tedious, and suffer from great individual variability.”

  • The paper relies on this without support.

    Accurate segmentation of most anatomical structures, particularly lesions in US images, is still challenging due to low contrast between the target and background.

    “Accurate segmentation of most anatomical structures, and particularly of lesions (nodules) in US images, is still a challenging task due to low contrast between the target and background in US images.”

  • The paper relies on this without support.

    Fully automatic left ventricle segmentation is challenging due to significant appearance and shape variations, low signal-to-noise ratio, shadows, and edge dropout.

    “However, fully automatic LV segmentation is a challenging task due to significant appearance and shape variations, a low signalto-noise ratio, shadows, and edge dropout.”

  • The paper relies on this without support.

    Fetal left ventricle segmentation is more challenging than adult left ventricle segmentation due to inhomogeneities, artifacts, poor contrast, large intersubject variations, and connected left ventricle and left atrium.

    “Compared with adult LV segmentation, fetal LV segmentation is more challenging, since fetal echocardiographic sequences suffer from inhomogeneities, artifacts, poor contrast, and large intersubject variations; furthermore, there is usually a connected LV and left atrium (LA) due to the random movements of the fetus in the womb.”

  • The paper relies on this without support.

    Boundary incompleteness is a common problem for many anatomical structures in medical US images and presents great challenges to automatic segmentation.

    “Boundary incompleteness is a common problem for many anatomical structures (e.g., prostate, breast, kidney, fetus, etc.) in medical US images, and presents great challenges to the automatic segmentation of these structures.”

  • The paper relies on this without support.

    US image analysis fields often struggle with limited training samples due to difficulty in generating and sharing lesion or disease images.

    “In contrast, US image analysis fields often struggle with limited training samples (usually in the hundreds or thousands, even after using data-augmentation strategies) due to the difficulty in generating and sharing lesion or disease images.”

  • The paper relies on this without support.

    Cross-domain transfer learning is the most common way to accomplish a variety of tasks in medical US analysis.

    “Cross-domain transfer learning is the most common way to accomplish a variety of tasks in medical US analysis.”

  • The paper relies on this without support.

    Pre-training of models is currently always performed on large sample datasets.

    “In any case, the pre-training of models is currently always performed on large sample datasets.”

  • The paper relies on this without support.

    Models trained on natural images may not be optimal for medical images, which are typically single channel, low contrast, and texture rich.

    “In addition, models trained on natural images may not be optimal for medical images, which are typically single channel, low contrast, and texture rich.”

  • The paper relies on this without support.

    Either US or mammography is usually regarded as the first-line screening examination, for which it is much easier to collect large training samples.

    “Either US or mammography (i.e., X-ray) is usually regarded as the first-line screening examination, for which it is much easier to collect large training samples.”

  • The paper relies on this without support.

    Breast MRI is more costly and time-consuming, commonly used for screening high-risk populations, and it is much more difficult to collect sufficient training datasets and ground-truth annotation for this method.

    “However, breast MRI is a more costly and timeconsuming method that is commonly used for screening highrisk populations, and it is much more difficult to collect sufficient training datasets and ground-truth annotation for this method.”

  • The paper relies on this without support.

    Current public availability of datasets in the field of medical US is still limited.

    “However, compared with the large and publicly available datasets in other areas (e.g., more than 1 million annotated multi-label natural images in ImageNet [6]), the current public availability of datasets in the field of medical US is still limited.”

  • The paper relies on this without support.

    Major performance improvement achievable with deep learning greatly depends on large training sample datasets.

    “Clearly, the major performance improvement that can be achieved with deep learning greatly depends on large training sample datasets.”

  • The paper relies on this without support.

    The effectiveness of any transfer method for a given target task mainly depends on two aspects: the source task and how it is related to the target.

    “Ignoring the inherent differences between different methods, the effectiveness of any transfer method for a given target task mainly depends on two aspects: the source task, and how it is related to the target.”

  • The paper relies on this without support.

    Producing positive transfer between sufficiently related tasks while avoiding negative transfer is difficult to achieve simultaneously in practice.

    “However, these goals are difficult to achieve simultaneously in practice.”

  • The paper relies on this without support.

    Mapping is necessary to translate between task representations when the representations of the source and target tasks are heterogeneous.

    “In addition, mapping is necessary in order to translate between task representations when the representations of the source and target tasks are heterogeneous.”

  • The paper relies on this without support.

    3D US is a very important imaging modality in medical imaging, and 3D US image analysis has shown great potential in US-based clinical application.

    “It is worth stressing again that 3D US is a very important imaging modality in the field of medical imaging, and that 3D US image analysis has shown great potential in US-based clinical application, although several issues remain to be addressed.”

  • The paper relies on this without support.

    It is currently difficult to proceed with the development of 3D deep learning methods without the strong support of other communities, especially the computer vision community.

    “However, it is currently difficult to proceed with the development of 3D deep learning methods without the strong support of other communities, and especially that of the CV community.”
20 of 20

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?

  • Since Hinton proposed the deep belief networks (DBN) in 2006, deep network-based methods have accomplished significant achievements in various sectors.

    “However, since Hinton proposed the deep belief networks (DBN) in 2006 [90], the deep network-based method has accomplished significant achievements in various sectors [91,92,93,94].”

    From Semantic Terrain Segmentation in the Navigation Vision of Planetary Rovers—A Systematic Literature Review (2022) · cites this paper as [93]

  • Deep learning is increasingly used in medical imaging for tasks such as automated lesion detection, segmentation and registration to assist clinicians in disease diagnosis and surgical planning.

    “In medical imaging, deep learning is increasingly used for tasks such as automated lesion detection, segmentation and registration to assist clinicians in disease diagnosis and surgical planning.”

    From Deep Learning Strategies for Ultrasound in Pregnancy (2020) · cites this paper as [9]

  • Deep learning techniques have the potential to create new screening tools, predict diseases, improve diagnostic accuracy and accelerate clinical tasks, whilst also reducing costs and human error.

    “Deep learning techniques have the potential to create new screening tools, predict diseases, improve diagnostic accuracy and accelerate clinical tasks, whilst also reducing costs and human error [9–17].”

    From Deep Learning Strategies for Ultrasound in Pregnancy (2020) · cites this paper as [9]

  • Accurate segmentation of most anatomical structures in medical ultrasound is challenging due to low contrast between the target and background of the images.

    “However, accurate segmentation of most anatomical structures in medical ultrasound is still challenging due to the low contrast between the target and background of the images [8].”

    From Model-Agnostic Method for Thoracic Wall Segmentation in Fetal Ultrasound Videos (2020) · cites this paper as [8]

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.

  • Caffe citation points to a fetal-ultrasound paper

    Minor

    The framework list cites reference 79 for Caffe, but reference 79 is about fetal ultrasound plane localization. The Caffe paper is listed as reference 80.

  • Jamieson citation identifies different authors and year

    Minor

    The text cites [86] for a 2012 study by Jamieson et al., but reference 86 is a 2008 paper by Giger et al. The Jamieson study appears as reference 87.

  • Fetal-plane citation points to a stroke study

    Minor

    Reference 115 is cited for traditional methods of detecting a fetal ultrasound standard plane, but its listed title concerns pathology following stroke in rats.

3 of 3

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Deng, 2014, Deep learning: methods and applications. Found Trends, Signal Process, 7, 197

Cited by

Sorted by citations

Related

Authors
Liu, S., Wang, Y., Yang, X., Lei, B., Liu, L., Li, S. X., Ni, D., & Wang, T.
Affiliations
  • Shenzhen University (CN)
  • Chinese University of Hong Kong (HK)
Corresponding author
  • Wang, Yi
  • Ni, Dong
  • Wang, Tianfu
ORCID
Venue
Engineering
Publisher
Elsevier BV
Published
2019-04
Volume
5
Issue
2
Pages
261-275
ISSN
2095-8099 (ISSN-L)
DOI
10.1016/j.eng.2018.11.020
OpenAlex
W2913559493
Semantic Scholar
0b706160de638668f72587b9df21f6361f57add4
Cited by
798
References
168
Citations by year
2026: 63 · 2025: 107 · 2024: 130 · 2023: 135 · 2022: 116 · 2021: 143 · 2020: 79 · 2019: 24 · 2018: 1
FWCI
40.02
Type
journal article
Language
en
Open access
Licence
Locations
  • Engineering (publishedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (61571304, 81571758, 61701312)
  • National Key Research and Development Program of China (2016YFC0104703)
  • Shenzhen Peacock Plan (KQTD2016053112051497)
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
  • Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
Keywords
ultrasonography, deep learning, image classification, object detection, image segmentation, deep learning architectures, diagnostic support

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 2026-10-10.

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

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