Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional Neural Networks and 3D Conditional Random Fields
Open accessThe DOI matches its Crossref record; 27 of the 27 references it cites were found in a registry or an index; 7 of the 19 sources its claims cite were checked for support.
Abstract
Automatic segmentation of the liver and its lesion is an important step towards deriving quantitative biomarkers for accurate clinical diagnosis and computer-aided decision support systems. This paper presents a method to automatically segment liver and lesions in CT abdomen images using cascaded fully convolutional neural networks (CFCNs) and dense 3D conditional random fields (CRFs). We train and cascade two FCNs for a combined segmentation of the liver and its lesions. In the first step, we train a FCN to segment the liver as ROI input for a second FCN. The second FCN solely segments lesions from the predicted liver ROIs of step 1. We refine the segmentations of the CFCN using a dense 3D CRF that accounts for both spatial coherence and appearance. CFCN models were trained in a 2-fold cross-validation on the abdominal CT dataset 3DIRCAD comprising 15 hepatic tumor volumes. Our results show that CFCN-based semantic liver and lesion segmentation achieves Dice scores over \(94\,\%\) for liver with computation times below 100 s per volume. We experimentally demonstrate the robustness of the proposed method as a decision support system with a high accuracy and speed for usage in daily clinical routine.
Overview
CoChat Summary
TL;DR
The paper proposes cascaded fully convolutional neural networks (CFCNs) followed by dense 3D conditional random fields (CRFs) for automatic liver and lesion segmentation in abdominal CT. On 15 hepatic tumor volumes from 3DIRCADb, the method reaches liver Dice of 94.3% and a lesion Dice of 82.3%, with computation times below 100s per volume.
Key findings
- Cascaded UNet + 3D CRF achieved liver Dice 94.3%, VOE 10.7%, RVD -1.4%, ASD 1.5 mm, and MSD 24.0 mm, improving over cascaded UNet alone (Dice 93.1%) and a single UNet (Dice 72.9%).
- Applying the 3D dense CRF improved slice-wise liver Dice from mean 93.1% to 94.3% (two-sided paired t-test p < 4·10⁻¹⁹).
- The cascade reduced lesion false positives by restricting lesion segmentation to the predicted liver ROI; the figure reports a lesion Dice score of 82.3% after 3D CRF.
- Runtime per slice was 2·0.2 s = 0.4 s without CRF and 0.8 s with CRF; the paper reports segmentation in under 100 s per volume.
Methods
- Evaluated on the 3DIRCADb dataset: 20 venous-phase CT volumes, using the 15 volumes containing hepatic tumors with 2-fold cross-validation; lesions per volume ranged from 1 to 42.
- Preprocessing windowed Hounsfield units to [-100, 400], applied histogram equalization, and augmented with translation, rotation, and Gaussian noise, yielding 22,693 image slices. Two cascaded UNets (19 convolutional layers) were trained: step 1 for liver, step 2 for lesions from the predicted liver ROI, with class-balancing weights in the cross-entropy loss.
- Trained in Caffe on an NVIDIA Titan X GPU at learning rate 0.001, momentum 0.8, and weight decay 0.0005. A dense 3D CRF with unary potentials from FCN probabilities and pairwise Potts/spatial/bilateral terms was applied via mean-field inference, with weights and kernels chosen by random search.
Limitations
- The paper does not state a dedicated limitations section.
- Evident from the text: the evaluation uses only 15 tumor volumes with 2-fold cross-validation, a small and heterogeneous sample.
- Evident from the text: lesion segmentation remains less accurate than liver segmentation (82.3% vs. 94.3% Dice), and the method uses 2D slice-wise FCNs rather than volumetric 3D convolutions because of GPU/VRAM limits and anisotropic resolution, relying on 3D CRF for volumetric context.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 1 Not valid · 6 Not checked · 18 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“In the first row, the false positive lesion prediction in B of a single UNet as proposed by [20] were eliminated in C by CFCN as a result of restricting lesion segmentation to the liver ROI region.”
What the sources say
- Supportedcites [20] 20. Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmentation. In: MICCAI, vol. 9351, pp. 234-241 (2015) source
The citation [20] is attached only to the identification 'a single UNet as proposed by [20]', not to the liver-ROI/false-positive result, which is the citing authors' own experiment and framing. Our copy of Ronneberger et al. 2015 plainly proposes that single end-to-end network: the Abstract (p. 1) presents 'a network and training strategy... a contracting path to capture context and a symmetric expanding path that enables precise localization... trained end-to-end', and Section 2 (Network Architecture, p. 4) describes it as one 23-convolutional-layer u-shaped network, with Introduction (p. 3) noting it 'does not have any fully connected layers'. Our copy contains no liver, lesion or CT content and no false-positive-lesion analysis (searched: liver lesion false positives region of interest), so nothing in it is contradicted by, or needed for, the citing sentence's attribution. (LLM, conf=0.85)
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Deep Convolutional Neural Networks CNN have gained new attention in the scientific community for solving computer vision tasks such as object recognition, classification and segmentation [14,18], often out-competing state-of-the art methods.”
What the sources say
- Not checked yetcites [14] 14. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: NIPS. pp. 1097-1105 (2012)
- Supportedcites [18] 18. Long, J., Shelhamer, E., Darrell, T.: Fully convolutional networks for semantic segmentation. CVPR (2015) source
The Abstract (p. 1) states that convolutional networks trained end-to-end, pixels-to-pixels, 'exceed the state-of-the-art in semantic segmentation', with state-of-the-art results on PASCAL VOC, NYUDv2 and SIFT Flow. The Introduction (p. 1) frames convnets as 'driving advances in recognition', improving for whole-image classification and making progress on object detection, part/keypoint prediction and correspondence — i.e. the same tasks the citing sentence lists (recognition, classification, segmentation). As a broad background citation this is what our copy clearly treats as the case; the only nuance is that explicit 'exceeds state-of-the-art' language is used for segmentation, while recognition/classification are described as advances rather than SOTA wins. (LLM, conf=0.80)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
None of the sources cited for this claim have been checked yet.
In the paper
“Most importantly, CNN methods have proven to be highly robust to varying image appearance, which motivates us to apply them to fully automatic liver and lesions segmentation in CT volumes.”
What the sources say
- Not checked yetcites [14] 14. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In: NIPS. pp. 1097-1105 (2012)
- Could not verifycites [18] 18. Long, J., Shelhamer, E., Darrell, T.: Fully convolutional networks for semantic segmentation. CVPR (2015) sourceExperimental — accuracy not yet measured
Our copy of Long, Shelhamer & Darrell (2015) never states that CNN/convnet methods are robust to varying image appearance. The Abstract (p. 1) calls convolutional networks 'powerful visual models that yield hierarchies of features' and the Introduction (p. 1) says convnets are 'driving advances in recognition'; the Results report state-of-the-art numbers on PASCAL VOC, NYUDv2 and SIFT Flow (Tables 1–6, pp. 7–9), but none of these passages frames the work as evidence of robustness to appearance variation. The only explicit robustness statement is the narrow Figure 6 caption (Results, p. 8) noting 'robustness to occluders' in sample outputs, which is a different property. Searches run: 'robust', 'robustness to varying image appearance', 'invariant features / transfer learning robustness', 'appearance information'. The paper also contains no medical/CT content, so it cannot be checked for the citing sentence's implied domain. (LLM, conf=0.68)
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Volumetric FCN implementation with 3D convolutions is strongly limited by GPU hardware and available VRAM [19].”
What the sources say
- Partly supportedcites [19] 19. Prasoon, A., Petersen, K., Igel, C., Lauze, F., Dam, E., Nielsen, M.: Deep feature learning for knee cartilage segmentation using a triplanar convolutional neural network. In: MICCAI. vol. 16, pp. 246-253 (2013) sourceExperimental — accuracy not yet measured
The Introduction says truly 3D CNNs have large memory and training-time requirements. Results, Discussion, and Future Work reports that a larger 3D CNN was computationally infeasible in the authors’ setup. The paper does not establish the claim specifically for volumetric FCNs or attribute the limit to GPU hardware and available VRAM. (LLM, conf=0.87)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The UNet architecture enables accurate pixel-wise prediction by combining spatial and contextual information in a network architecture comprising 19 convolutional layers.”
What the sources say
- Partly supportedcites [20] 20. Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical image segmentation. In: MICCAI, vol. 9351, pp. 234-241 (2015) sourceExperimental — accuracy not yet measured
The qualitative part of the claim is well supported: the Abstract (p. 1) describes 'a contracting path to capture context and a symmetric expanding path that enables precise localization', and the Introduction (p. 3) says the large number of feature channels in the upsampling part 'allow the network to propagate context information to higher resolution layers', with pixel-wise (per-pixel) output established in the Introduction (pp. 1-2) and the pixel-wise soft-max in Training (p. 4). However, the layer count attributed here does not match: the Network Architecture section (p. 4) states 'In total the network has 23 convolutional layers', not 19. That numeric qualifier is therefore unsupported (indeed contrary to our copy). (LLM, conf=0.85)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
18 more claims · 2 with their cited sources checked
Create a free account to see all 30 claims and to verify the 12 references still open.
- Probably not a substantial claim. 4
- It may state two separate claims as one. 4
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Because the second type of 3D U-net focused on the segmentation of liver mass, the segmentation of liver mass was more efficiently performed than single 3D U-net.”
From Convolutional neural networks: an overview and application in radiology (2018) · cites this paper as [9]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Another study [9] utilized two types of 3D U-net for segmenting liver and liver mass on 3D CT images, which was named cascaded fully convolutional neural networks; one type of U-net was used for segmentation of the liver and the other type for the segmentation of liver mass using the liver segmentation results.”
From Convolutional neural networks: an overview and application in radiology (2018) · cites this paper as [9]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“several studies have already been published in areas such as lesion detection [7], classification [8], segmentation [9], image reconstruction [10, 11], and natural language processing [12].”
From Convolutional neural networks: an overview and application in radiology (2018) · cites this paper as [9]
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
U-Net: Convolutional Networks for Biomedical Image Segmentation
Ronneberger, O., Fischer, P., & Brox, T. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-24574-4_28 · Cited by 99,404
Fully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834
Fully convolutional networks for semantic segmentation
Shelhamer, E., Long, J., & Darrell, T. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2572683 · Cited by 11,898
Caffe: Convolutional Architecture for Fast Feature Embedding
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., & Darrell, T. (2014) · Proceedings of the 22nd ACM international conference on Multimedia · doi:10.1145/2647868.2654889 · Cited by 11,077
Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs
Chen, L.-C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2014) · arXiv · doi:10.48550/arxiv.1412.7062 · Cited by 4,213 · cited 1×
“Semantic image segmentation methods based on fully convolutional neural networks FCN were developed in [18], with impressive results in natural image segmentation competitions [3,24].”
MethodCaffe: Convolutional Architecture for Fast Feature Embedding
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., & Darrell, T. J. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1408.5093 · Cited by 3,752
Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation
Kamnitsas, K., Ledig, C., Newcombe, V. F. J., Simpson, J. P., Kane, A. D., Menon, D. K., Rueckert, D., & Glocker, B. (2017) · Medical Image Analysis · doi:10.1016/j.media.2016.10.004 · Cited by 3,498 · cited 1×
“[20,8,23,21,19,12].”
Brain tumor segmentation with Deep Neural Networks
Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C., Jodoin, P.-M., & Larochelle, H. (2017) · Medical Image Analysis · doi:10.1016/j.media.2016.05.004 · Cited by 3,278
Brain Tumor Segmentation with Deep Neural Networks
Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., Pal, C., Jodoin, P.-M., & Larochelle, H. (2015) · arXiv · doi:10.48550/arxiv.1505.03540 · Cited by 3,278
Research works printed but not linked: 1
- 4. Deng, X., Du, G.: Editorial: 3d segmentation in the clinic: a grand challenge ii-liver tumor segmentation. In: MICCAI Workshop (2008)
Cited by
A survey on deep learning in medical image analysis
Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A. A., Ciompi, F., Ghafoorian, M., van der Laak, J. A. W. M., van Ginneken, B., & Sánchez, C. I. (2017) · Medical Image Analysis · doi:10.1016/j.media.2017.07.005 · Cited by 15,332
Convolutional neural networks: an overview and application in radiology
Yamashita, R., Nishio, M., Do, R. K. G., & Togashi, K. (2018) · Insights into Imaging · doi:10.1007/s13244-018-0639-9 · Cited by 4,682
“Needless to say, there has been a surge of interest in the potential of CNN among radiology researchers, and several studies have already been published in areas such as lesion detection [7], classification [8], segmentation [9], image reconstruction [10, 11], and natural language processing [12].”
BackgroundMethodH-DenseUNet: Hybrid Densely Connected UNet for Liver and Tumor Segmentation From CT Volumes
Li, X., Chen, H., Qi, X., Dou, Q., Fu, C.-W., & Heng, P.-A. (2018) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2018.2845918 · Cited by 2,473
MultiResUNet : Rethinking the U-Net architecture for multimodal biomedical image segmentation
Ibtehaz, N., & Rahman, M. S. (2020) · Neural Networks · doi:10.1016/j.neunet.2019.08.025 · Cited by 2,311
Deep Learning Techniques for Medical Image Segmentation: Achievements and Challenges
Hesamian, M. H., Jia, W., He, X., & Kennedy, P. (2019) · Journal of Digital Imaging · doi:10.1007/s10278-019-00227-x · Cited by 1,631
Metadata
- Authors
- Christ, P. F., Elshaer, M. E. A., Ettlinger, F., Tatavarty, S., Bickel, M., Bilic, P., Rempfler, M., Armbruster, M., Hofmann, F., D’Anastasi, M., Sommer, W. H., Ahmadi, S.-A., & Menze, B. H.
- Affiliations
- Technical University of Munich (DE)
- LMU Klinikum (DE)
- Corresponding author
- Christ, Patrick Ferdinand
- Venue
- Lecture Notes in Computer Science
- Publisher
- Springer International Publishing
- Published
- 2016
- Pages
- 415-423
- ISSN
- 0302-9743 (ISSN-L)
- 1611-3349
- OpenAlex
- W2526009326
- Semantic Scholar
- 59d5a686c82c9c73ee5f6ef42b8be0c245dc84b1
- Cited by
- 625
- References
- 25
- Citations by year
- 2026: 14 · 2025: 33 · 2024: 54 · 2023: 49 · 2022: 77 · 2021: 116 · 2020: 76 · 2019: 100 · 2018: 73 · 2017: 30 · 2016: 2
- FWCI
- 23.71
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springer.com/tdm (tdm)
- www.springer.com/tdm (vor)
- Locations
- Lecture notes in computer science (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Engineering › Biomedical Engineering › Medical Imaging and Analysis
- Keywords
- liver segmentation, liver lesion segmentation, abdominal CT, semantic segmentation, liver tumors, computer-aided diagnosis, decision support system
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional Neural Networks and 3D Conditional Random Fields. https://app.cochat.ai/papers/10.1007/978-3-319-46723-8_48
- The paper
- Christ, P. F., Elshaer, M. E. A., Ettlinger, F., Tatavarty, S., Bickel, M., Bilic, P., Rempfler, M., Armbruster, M., Hofmann, F., D’Anastasi, M., Sommer, W. H., Ahmadi, S.-A., & Menze, B. H. (2016). Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional Neural Networks and 3D Conditional Random Fields. Lecture Notes in Computer Science. https://doi.org/10.1007/978-3-319-46723-8_48
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.