Comparative performance of 3D-DenseNet, 3D-ResNet, and 3D-VGG models in polyp detection for CT colonography
Abstract
Three-dimensional (3D) convolutional neural networks (CNNs) can process volumetric medical imaging data in their native volumetric input form. However, there is little information about the comparative performance of such models in medical imaging in general and in CT colonography (CTC) in particular. We compared the performance of a 3D densely connected CNN (3D-DenseNet) with those of the popular 3D residual CNN (3D-ResNet) and 3D Visual Geometry Group CNN (3D-VGG) in the reduction of false-positive detections (FPs) in computer-aided detection (CADe) of polyps in CTC. VGG is the earliest CNN design of these three models. ResNet has been used widely as a de-facto standard model for constructing deep CNNs for image classification in medical imaging. DenseNet is the most recent of these models and improves the flow of information and reduces the number of network parameters as compared to those of ResNet and VGG. For the evaluation, we used 403 CTC datasets from 203 patients. The classification performance of the CNNs was evaluated by use of 5-fold cross-validation, where the area under the receiver operating characteristic curve (AUC) was used as the figure of merit. Each training fold was balanced by use of data augmentation of the samples of real polyps. Our preliminary results showed that the AUC value of the 3D-DenseNet (0.951) was statistically significantly higher than those of the reference models (P < 0.005), indicating that the 3D-DenseNet has the potential of substantially outperforming the other models in reducing FPs in CADe for CTC. This improvement was highest for the smallest polyps.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
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 3D ResNet50 architecture has proven effective in handling 3D medical imaging data.
“we opted for its 3D counterpart, which has proven effective in handling 3D medical imaging data46.”
From Foundation Model for Cancer Imaging Biomarkers (2024) · cites this paper as [46]
Cited by
Sorted by citations
Foundation Model for Cancer Imaging Biomarkers
Pai, S., Bontempi, D., Hadzic, I., Prudente, V., Sokač, M., Chaunzwa, T. L., Bernatz, S., Hosny, A., Mak, R. H., Birkbak, N. J., et al. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-024-00807-9 · Cited by 199 · see citing works
“While the SimCLR authors used a 2D ResNet50 architecture, we opted for its 3D counterpart, which has proven effective in handling 3D medical imaging data 46 .”
Pellicer-Valero, Ó. J., Jiménez, J. L. M., González‐Pérez, V., Ramón-Borja, J. L. C., García, I. M., Benito, M. B., Gómez, P. P., Rubio‐Briones, J., Rupérez, M. J., & Martín‐Guerrero, J. D. (2022) · Scientific Reports · doi:10.1038/s41598-022-06730-6 · Cited by 116 · see citing works
“In this sense, CAD is one of the most exciting lines of research in medical imaging and has been successfully applied to interpret images in different medical scenarios 40,18,68,64 .”
BackgroundMehta, N. K., Prasad, S. S., Saurav, S., Saini, R., & Singh, S. (2022) · Applied Intelligence · doi:10.1007/s10489-022-03200-4 · Cited by 97 · see citing works
“Since the development of C3D neural networks [35], several deep 3D CNNs have been proposed for numerous computer vision applications [36].”
BackgroundWang, L., Ding, W., Mo, Y., Shi, D., Zhang, S., Zhong, L., Wang, K., Wang, J., Huang, C., Zhang, S., et al. (2021) · European Journal of Nuclear Medicine and Molecular Imaging · doi:10.1007/s00259-021-05432-x · Cited by 59 · see citing works
“…3D-ResNet achieved surprising results in various medical scenarios, such as recessive ischemic stroke recognition [25], Alzheimer’s disease classification [31], thoracic vertebral segmentation [32], and colo-noscopic polyp detection [33], with an accuracy range of 0.83 from 0.97 in the test set.”
BackgroundInfluentialYang, Q., Yu, D., Zhang, Z., et al. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.3046240 · Cited by 54 · see citing works
Folle, L., Bayat, S., Kleyer, A., Fagni, F., Kapsner, L. A., Schlereth, M., Meinderink, T., Breininger, K., Taşçılar, K., Krönke, G., et al. (2022) · Lara D. Veeken · doi:10.1093/rheumatology/keac197 · Cited by 51 · see citing works
Rapid pavement aggregate gradation estimation based on 3D data using a multi-feature fusion network
Weng, Z., Ablat, G., Wu, D., et al. (2021) · Automation in Construction · doi:10.1016/j.autcon.2021.104050 · Cited by 46 · see citing works
Oh, S., Kang, S., Oh, I., & Kim, M.-S. (2023) · BMC Bioinformatics · doi:10.1186/s12859-023-05160-z · Cited by 27 · see citing works
Uemura, T., Näppi, J. J., Ryu, Y., Watari, C., Kamiya, T., & Yoshida, H. (2020) · International Journal of Computer Assisted Radiology and Surgery · doi:10.1007/s11548-020-02275-z · Cited by 15 · see citing works
“We trained our previously developed 3DDenseNetCNN [30] to detect polyps with five data augmentation approaches: baseline augmentation, nonlinear augmentation, 3D GAN with baseline augmentation, 3D Glow with baseline augmentation and 3D Glow with nonlinear augmentation.”
MethodSelf-citationComparative Analysis of Deep Learning Approaches for the Diagnosis of Diabetic Retinopathy
Gulati, S., Singh, V. P., & Shukla, S. (2022) · Students Conference on Engineering and Systems · doi:10.1109/sces55490.2022.9887778 · Cited by 12 · see citing works
“In this paper, we have used three CNN models for the comparative analysis [20]: 1) ResNet50: ResNet is a short form for residual network and this architecture uses convolutional neural network multiple times.”
MethodWei, C., Liu, Z., Zhang, Y., & Fan, L. (2025) · Digital Health · doi:10.1177/20552076251314546 · Cited by 10 · see citing works
“10 Various early diagnostic systems that assist radiologists in accurate lesion segmentation have been proposed, marking the inception of this fi eld.”
He, S., Li, Q., Li, X., et al. (2023) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2023.105275 · Cited by 9 · see citing works
Foundation Models for Quantitative Biomarker Discovery in Cancer ImagingPreprint
Pai, S., Bontempi, D., Hadžić, I., Prudente, V., Sokač, M., Chaunzwa, T. L., Bernatz, S., Hosny, A., Mak, R. H., Birkbak, N. J., et al. (2023) · medRxiv · doi:10.1101/2023.09.04.23294952 · Cited by 9 · see citing works
Altal, O. F., Sindiani, A., Mhanna, H. Y. A., et al. (2025) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2025.1664201 · Cited by 6 · see citing works
Multiple instance ensembling for paranasal anomaly classification in the maxillary sinus
Bhattacharya, D., Behrendt, F., Becker, B., Beyersdorff, D., Petersen, E. L., Petersen, M., Cheng, B., Eggert, D., Betz, C. S., Hoffmann, A. S., et al. (2023) · International Journal of Computer Assisted Radiology and Surgery · doi:10.1007/s11548-023-02990-3 · Cited by 6 · see citing works
Beigi, M. A., & Ghazizadeh-Ahsaee, M. (2025) · SN Computer Science · doi:10.1007/s42979-024-03569-8 · Cited by 6 · see citing works
“VGG is one of the earliest CNN-based models developed for image classification problems, and ResNet is one of the architectures widely used for medical image classification [ 24 ].”
MRI Liver Image Assisted Diagnosis Based on Improved Faster R-CNN
Tao, M., Lou, J., & Wang, L. (2022) · Traitement du signal · doi:10.18280/ts.390428 · Cited by 5 · see citing works
“Some scholars have demonstrated that for small target detection in medicine, if DenseNet is employed as a feature extraction network for Faster R-CNN, the experimental performance outperforms the VGG16 as well as the ResNet structure employed in the original Faster R-CNN [30].”
MethodLVCL: Label-Volume Contrastive Learning for Multi-Label Classification of Retinal Oct Volumes
Zhang, Y., Tang, Z., Yang, D., et al. (2023) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi53787.2023.10230396 · Cited by 4 · see citing works
Colorectal Polyps Detection in Virtual Colonoscopy Using 3D Geometric Features and Deep Learning
Yousuf, M., Harb, S., Alkabbany, I. M., et al. (2024) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi56570.2024.10635785 · Cited by 4 · see citing works
“Ue-mura and N¨appi [9] proposed a method to classify polyps using 3D CNN, however, the network needs high computational power in order to be trained on 3D data as shown in the experiments section.”
MethodLuximon, D. C., Neylon, J., & Lamb, J. M. (2023) · Physics and Imaging in Radiation Oncology · doi:10.1016/j.phro.2023.100427 · Cited by 3 · see citing works
“A 3D Dense-Net [24] may potentially improve the performance of the ARL by obtaining more useful features as compared to the current 2D model [25].”
MethodNako, S., Seo, C., Tsutsui, S., et al. (2023) · Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society · doi:10.1109/embc40787.2023.10340345 · Cited by 1 · see citing works
“After the annotation of each created 3D data with thrombus presence or absence label, the annotated 3D dataset is inputted to the 3D ResNet [7] so that classification learning of thrombus presence/absence is performed.”
MethodAction Recognition with 3D Residual Attention and Cross Entropy
Ouyang, Y., & Li, X. (2025) · Entropy · doi:10.3390/e27040368 · Cited by 2 · see citing works
“A more effective solution is to use 3D CNNs to extract spatiotemporal details from stacked RGB frames within a continuous deep network framework [9,10].”
Li, S., Wang, C., Li, F., et al. (2026) · npj Precision Oncology · doi:10.1038/s41698-026-01315-2 · Cited by 1 · see citing works
An Automatic Colorectal Polyps Detection Approach for Ct Colonography
Yousuf, M., Alkabbany, I. M., Ali, A., et al. (2023) · Proceedings - International Conference on Image Processing · doi:10.1109/icip49359.2023.10221981 · Cited by 1 · see citing works
“Uemura and N¨appi’s work [17] shows a way to classify polyps using 3D networks, but it needs high computational power in order to feed the CNN network with 3D data alongside with high-time computation needed due to large data size.”
BackgroundMethodMulti-View Network for Colorectal Polyps Detection in CT Colonography
Yousuf, M., Harb, S., Alkabbany, I. M., Ali, A., Elshazley, S., & Farag, A. A. (2024) · Proceedings - International Conference on Image Processing · doi:10.1109/icip51287.2024.10647279
Loose Lesion Location Self-supervision Enhanced Colorectal Cancer Diagnosis
Gao, T., Song, J., Yu, X., et al. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-72120-5_39
Enhancing Skin Cancer Detection Through Category Representation and Fusion of Pre-Trained ModelsPreprint
Kong, L., Velasquez, J. D., Snåšel, V., et al. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5119282
Accelerating CNN Inference: Hybrid Model Pruning for Reduced Computational Demands and High Accuracy
Gadah, A. R., Isaak, S., & Saad, A. H. Y. (2024) · IEEE Student Conference on Research and Development (SCOReD) · doi:10.1109/scored64708.2024.10872661
“AlexNet is one of the earliest well-known CNNs that pioneered the use of deep architecture in large-scale image recognition tasks [11]; history has proven that as more advanced architectures such as VGGNet, ResNet, and the significantly improved EfficientNet architectures increase the performance…”
Multi-feature colon polyp detection algorithm based on dual-position CT images
Qiu, S., Lu, H., Shu, J., et al. (2025) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2025.109209
BRIQA: Balanced Reweighting in Image Quality Assessment of Pediatric Brain MRI
Almsouti, A., Khamitova, A., Taratynova, D., et al. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-14417-1_1
- Authors
- Uemura, T., Näppi, J. J., Hironaka, T., Kim, H., & Yoshida, H.
- Affiliations
- Harvard University (US)
- Kyushu Institute of Technology (JP)
- Venue
- Medical Imaging 2020: Computer-Aided Diagnosis
- Publisher
- SPIE
- Published
- 2020-03-16
- Pages
- 110
- OpenAlex
- W3011008306
- Cited by
- 31
- References
- 0
- Citations by year
- 2026: 3 · 2025: 5 · 2024: 5 · 2023: 9 · 2022: 5 · 2021: 2 · 2020: 2
- FWCI
- 8.09
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- SPIE Medical Imaging: Computer-Aided Diagnosis (publishedVersion)
- Topics
- Health Sciences › Medicine › Oncology › Colorectal Cancer Screening and Detection
- Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- CT colonography, polyp detection, computer-aided detection, 3D convolutional neural networks, 3D DenseNet, 3D ResNet, false positive reduction, volumetric medical imaging, data augmentation
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.
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
