BREAST LESION CLASSIFICATION: LEAKAGE-FREE PATCH-BASED TRANSFER LEARNING WITH CROSS- DATASET EVALUATION ON CBIS-DDSM, MIAS AND INBREAST

(Received: 3-Jul.-2026, Revised: 23-Aug.-2026 and 27-Aug.-2026 , Accepted: 28-Aug.-2026)

Authors Hitesh Kag, Dinesh Kumar, Anjali Diwan,

Keywords #Breast lesion classification #Data leakage #Transfer learning #External validation #Mammography

Abstract High reported accuracies on small mammographic datasets are frequently inflated by data leakage, most often from augmentation applied before the train/test split. Changing only the split order on whole-image Mini-MIAS moves a deep-feature plus Support Vector Machine (SVM) pipeline from 0.976 accuracy to 0.513, which is equivalent to the random chance. The predictive signal survives at the lesion-patch scale. Our contribution is protocol-level rather than architectural: a reproducible leakage audit; the first quantification and repair of the reported patient overlap in the CBIS-DDSM official split, which we trace to its mass and calcification partitions being drawn independently (31 patients, 8.5% of test lesions); an architecture sweep that tests rather than assumes the performance ceiling; and a label-free procedure for transferring the decision threshold to a new institution. A lightweight EfficientNet patch pipeline trained on CBIS-DDSM under a strict patient-wise protocol obtains ROC- AUC 0.745 (95% CI 0.703–0.794); without retraining, it attains 0.734 (0.640–0.826) on MIAS and 0.877 (0.804– 0.946) on INbreast and a DeLong test cannot distinguish the MIAS result from the internal one (p = 0.82). Results are stable over ten seeds (σ≤ 0.011) with clustered bootstrap intervals. Across 11 configurations spanning six architecture families, test AUC spans only 0.707–0.777. For small datasets, leakage control and external validation matter more than headline accuracy.

References

[1] A. Bhushan, A. Gonsalves and J. U. Menon, "Current State of Breast Cancer Diagnosis, Treatment andTheranostics," Pharmaceutics, vol. 13, no. 5, p. 723, 2021.

[2] American Cancer Society, "Limitations of Mammograms," [Online], Available:https://www.cancer.org/cancer/types/breast-cancer/screening-tests-and-early-detection/mammograms/limitations-of-mammograms.html, Jun. 2023.

[3] L. Luo et al., "Deep Learning in Breast Cancer Imaging: A Decade of Progress and Future Directions,"IEEE Reviews in Biomedical Engineering, 2024, DOI: 10.1109/RBME.2024.3357567.

[4] H. E. Kim et al., "Transfer Learning for Medical Image Classification: A Literature Review," BMCMedical Imaging, vol. 22, p. 69, DOI: 10.1186/s12880-022-00793-7, 2022.

[5] D. Lévy and A. Jain, "Breast Mass Classification from Mammograms Using Deep Convolutional NeuralNetworks," arXiv Preprint, 1612.00542, [Online], Available: http://arxiv.org/abs/1612.00542, 2016.

[6] D. A. Ragab, M. Sharkas, S. Marshall and J. Ren, "Breast Cancer Detection Using Deep ConvolutionalNeural Networks and Support Vector Machines," PeerJ, vol. 7, p. e6201, DOI: 10.7717/peerj.6201, 2019.

[7] A. S. Setiawan et al., "Mammogram Classification Using Law’s Texture Energy Measure and Neural Networks," Procedia Computer Science, vol. 59, pp. 92–97, DOI: 10.1016/j.procs.2015.07.341, 2015.

[8] S. A. Agnes et al., "Classification of Mammogram Images Using Multiscale All Convolutional NeuralNetwork (MA-CNN)," J. of Medical Systems, vol. 44, p. 30, DOI: 10.1007/s10916-019-1494-z, 2020.

[9] S. Kapoor and A. Narayanan, "Leakage and the Reproducibility Crisis in Machine-learning-basedScience," Patterns, vol. 4, no. 9, p. 100804, DOI: 10.1016/j.patter.2023.100804, 2023.

[10] A. Lemay et al., "Improving the Repeatability of Deep Learning Models with Monte Carlo Dropout,"NPJ Digital Medicine, vol. 5, p. 174, DOI: 10.1038/s41746-022-00709-3, 2022.

[11] Y. Yang et al., "MammoDG: Generalizable Deep Learning Breaks the Limits of Cross-domain Multi-center Breast Cancer Screening," arXiv preprint, arXiv: 2308.01057, 2023.

[12] Y. Zafari et al., "MammoClean: Toward Reproducible and Bias-aware AI in Mammography throughDataset Harmonization," arXiv preprint, arXiv Preprint, arXiv: 2511.02400, Nov. 2025.

[13] S. Benbakreti et al., "Using ResNet-18 in a Deep-learning Framework and Assessing the Effects ofAdaptive Learning rates in the Identification of Malignant Masses in Mammograms," Jordanian Journal of Computers and Information Technology (JJCIT), vol. 10, no. 1, pp. 93–107, DOI: 10.5455/jjcit.71-1699818406, 2024.

[14] X. Yu, Z. Ren, D. S. Guttery and Y.-D. Zhang, "DF-dRVFL: A Novel Deep Feature Based Classifier forBreast Mass Classification," Multimedia Tools and Applications, vol. 83, no. 5, pp. 14393-14422, 2023.

[15] R. S. Lee et al., "A Curated Mammography Data Set for Use in Computer-aided Detection and DiagnosisResearch," Scientific Data, vol. 4, p. 170177, DOI: 10.1038/sdata.2017.177, 2017.

[16] I. C. Moreira et al., "INbreast: Toward a Full-field Digital Mammographic Database," AcademicRadiology, vol. 19, no. 2, pp. 236– 248, DOI: 10.1016/j.acra.2011.09.014, 2012.

[17] H. T. Nguyen et al., "VinDr-Mammo: A Large-scale Benchmark Dataset for Computer-aided Diagnosisin Full-field Digital Mammography," Scientific Data, vol. 10, p. 277, DOI: 10.1038/s41597-023- 02100-7, 2023.

[18] G. I. Quintana et al., "Exploiting Patch Sizes and Resolutions for Multi-scale Deep Learning inMammogram Image Classification," Bioengineering, vol. 10, no. 5, p. 534, DOI: 10.3390/bioengineering10050534, 2023.

[19] P. Oza, P. Sharma, S. Patel, F. Adedoyin and A. Bruno, "Image Augmentation Techniques forMammogram Analysis," Journal of Imaging, vol. 8, no. 5, p. 141, DOI: 10.3390/jimaging8050141, 2022.

[20] A. P. Adedigba, S. A. Adeshina and A. M. Aibinu, "Performance Evaluation of Deep Learning Modelson Mammogram Classification Using Small Dataset," Bioengineering, vol. 9, no. 4, p. 161, 2022.

[21] A. G. M. Al Mansour, F. Alshomrani, A. Alfahaid and A. T. M. Almutairi, "MammoViT: A CustomVision Transformer Architecture for Accurate BI-RADS Classification in Mammogram Analysis," Diagnostics, vol. 15, no. 3, p. 285, DOI: 10.3390/diagnostics15030285, 2025.

[22] S. Elumalai et al., "Breast Cancer Classification Based on Micro-calcifications Using Dual Branch VisionTransformer Fusion," Scientific Reports, vol. 15, DOI: 10.1038/s41598- 025-34377-6, 2025.

[23] S. Sharma, Y. Singh and T. Choudhury, "Advanced Deep Learning Architectures for EnhancedMammography Classification: A Comparative Study of CNNs and ViT," Discover Artificial Intelligence, vol. 5, Article no. 187, DOI: 10.1007/s44163-025-00426-2, 2025.

[24] S. Ghosh et al., "Mammo-CLIP: A Vision Language Foundation Model to Enhance Data Efficiency andRobustness in Mammography," Proc. Med. Image Comput. Comput.-Assist. Interv. (MICCAI 2024), vol. LNCS 15012, DOI: 10.1007/978-3-031-72390-2_59, Morocco, 2024.

[25] F. M. Talaat et al., "Grad-CAM Enabled Breast Cancer Classification with a 3D Inception-ResNet V2:Empowering Radiologists with Explainable Insights," Cancers, vol. 16, no. 21, p. 3668, 2024.

[26] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad-CAM: VisualExplanations from Deep Networks via Gradient-based Localization," Proc. of the IEEE Int. Conf. on Computer Vision (ICCV), pp. 618–626, Venice, Italy, 2017.

[27] N. Shifa, M. Saleh, Y. Akbari and S. Al Maadeed, "A Review of Explainable AI Techniques and TheirEvaluation in Mammography for Breast Cancer Screening," Clinical Imaging, vol. 123, p. 110492, 2025.

[28] H. C. Lu, E. W. Loh and S. C. Huang, "The Classification of Mammogram Using Convolutional NeuralNetwork with Specific Image Pre-processing for Breast Cancer Detection," Proc. of the 2nd IEEE Int. Conf. on Artificial Intelligence and Big Data (ICAIBD), pp. 9–12, Chengdu, China, 2019.

[29] M. Tan and Q. V. Le, "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,"Proc. of the 36th Int. Conf. on Machine Learning (ICML), pp. 6105–6114, PMLR 97, L. B., California, USA, 2019.

[30] S. R. Salian and S. D. Sawarkar, "Melanoma Skin Lesion Classification Using Improved Efficient-NetB3," Jordanian Journal of Computers and Information Technology (JJCIT), vol. 8, no. 1, pp. 45–56, DOI: 10.5455/jjcit.71-1636005929, 2022.

[31] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," Proc. of the 2016IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2016.

[32] G. Huang et al., "Densely Connected Convolutional Networks," Proc. of the IEEE Conf. on ComputerVision and Pattern Recognition (CVPR), pp. 4700–4708, 2017.[33] Z. Liu et al., "A ConvNet for the 2020s," Proc. of the 2022 IEEE/CVF Conf. on Computer Vision andPattern Recognition (CVPR), pp. 11976–11986, 2022.

[34] A. Dosovitskiy et al., "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,"arXiv Preprint, arXiv: 2010.11929, 2021.

[35] Z. Liu et al., "Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows," Proc. of the2021 IEEE/CVF Int. Conf. on Computer Vision (ICCV), pp. 9992–10002, 2021.

[36] E. R. DeLong et al., "Comparing the Areas under Two or More Correlated Receiver OperatingCharacteristic Curves: A Nonparametric Approach," Biometrics, vol. 44, no. 3, pp. 837–845, 1988.

[37] X. Sun and W. Xu, "Fast Implementation of DeLong’s Algorithm for Comparing the Areas underCorrelated Receiver Operating Characteristic Curves," IEEE Signal Process. Lett., vol. 21, no. 11, pp. 1389–1393, 2014.

[38] C. Guo, G. Pleiss, Y. Sun and K. Q. Weinberger, "On Calibration of Modern Neural Networks,"Proceedings of the 34th International Conference on Machine Learning (ICML'17) – vol. 70, pp. 1321 – 1330, 2017.

[39] W. J. Youden, "Index for Rating Diagnostic Tests," Cancer, vol. 3, no. 1, pp. 32–35, 1950.

[40] W. M. Salama and M. H. Aly, "Deep Learning in Mammography Images Segmentation andClassification: Automated CNN Approach," Alexandria Engineering Journal, vol. 60, no. 5, pp. 4701–4709, 2021.

[41] Z. Jafari and E. Karami, "Breast Cancer Detection in Mammography Images: A CNN-based Approachwith Feature Selection," Information, vol. 14, no. 7, DOI: 10.3390/info14070410, 2023.