Abstract

In this study, an explainable deep learning model combining EfficientNetB0 and a Prototype Layer (ProtoPNet) was developed for the classification of skin lesions. The model assigned two prototypes per class to learn specific class features and employed both Grad-CAM and a novel Most Similar Patch method to explain decision mechanisms following prediction. During Most Similar Patch analysis, LIME contour maps were generated for patches with high edge density, whereas color histograms were used for regions characterized predominantly by color and texture. The model’s ability to achieve both accurate classification and causal reasoning was evaluated, showing promising results in learning intra- class representations and providing meaningful explanations. Future work envisions extending this approach to more complex datasets such as histopathological images, enabling reliable interpretation of critical structures like tumor microenvironments. This hybrid framework is expected to contribute to the development of clinically trustworthy, explainable AI systems applicable to medical image analysis

References

[1] Hamim SA, Tamim MUI, Mridha MF, Safran M, Che D. SmartSkin-XAI: An interpretable deep learning approach for enhanced skin cancer diagnosis in smart healthcare. Diagnostics 2025;15:64. https://doi.org/10.3390/diagnostics15010064.

[2] Mondol RK, Millar EKA, Graham PH, Browne L, Sowmya A, Meijering E. GRAPHite: Graph- based interpretable tissue examination for enhanced explainability in breast cancer histopathology. arXiv preprint 2025;arXiv:2501.04206v1. https://doi.org/10.48550/arXiv.2501.04206.

[3] Donmez TB, Kutlu M, Mansour M, Yildiz MZ. Explainable AI in action: a comparative analysis of hypertension risk factors using SHAP and LIME. Neural Comput Appl 2025;37:4053– 4074. https://doi.org/10.1007/s00521-024-10724-y.

[4] Altini N, Brunetti A, Puro E, Taccogna MG, Saponaro C, Zito FA, De Summa S, Bevilacqua

V. NDG-CAM: Nuclei detection in histopathology images with semantic segmentation networks and Grad-CAM. Bioengineering 2022;9:475. https://doi.org/10.3390/bioengineering9090475.

[5] Mondol RK, Millar EKA, Graham PH, Browne L, Sowmya A, Meijering E. hist2RNA: An efficient deep learning model to predict gene expression from breast cancer histopathology images. Cancers 2023;15:2569. https://doi.org/10.3390/cancers15092569.

[6] Chiaburu T, Bießmann F, Haußer F. Uncertainty propagation in XAI: A comparison of analytical and empirical estimators. arXiv preprint 2025;arXiv:2504.03736v1. https://arxiv.org/abs/2504.03736.

[7] Manz R, Bäcker J, Cramer S, Meyer P, Müller D, Muzalyova A, et al. Do explainable AI (XAI) methods improve the acceptance of AI in clinical practice? An evaluation of XAI methods on Gleason grading. J Pathol Clin Res 2025;11:e70023. https://doi.org/10.1002/2056-4538.70023.

[8] Özkurt C. Advancing skin cancer diagnosis through the comparison of SHAP and Layer-wise Relevance Propagation (LRP). Research Square 2024. https://doi.org/10.21203/rs.3.rs- 3920847/v1.

[9] Slack D, Hilgard S, Singh S, Lakkaraju H. Reliable post hoc explanations: Modeling uncertainty in explainability. arXiv preprint 2021;arXiv:2008.05030. http://arxiv.org/abs/2008.05030.

[10] Evans T, Retzlaff CO, Geißler C, et al. The explainability paradox: challenges for XAI in digital pathology. Future Gener Comput Syst 2022;133:281–296. 

[11] Tjoa E, Guan C. A survey on explainable artificial intelligence (XAI): Towards medical XAI. IEEE Trans Neural Netw Learn Syst 2020;32(11):4793–4813. https://doi.org/10.1109/TNNLS.2020.3027314.

[12] Ribeiro MT, Singh S, Guestrin C. "Why should I trust you?": Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; 2016. p. 1135–1144.https://doi.org/10.1145/2939672.2939778.

[13] Ribera M, Lapedriza A. Can we do better explanations? A proposal of User-Centered Explainable AI. In: Joint Proceedings of the ACM IUI 2019 Workshops; 2019; Los Angeles, USA. ACM: New York, NY. http://hdl.handle.net/10609/99643.

[14] Codella N, Rotemberg V, Tschandl P, et al. Skin lesion analysis toward melanoma detection: A challenge at the 2018 ISIC dermoscopic image analysis workshop. arXiv preprint 2018;arXiv:1902.03368. https://doi.org/10.48550/arXiv.1902.03368 (accessed Dec 26, 2024).

[15] Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D. Grad-CAM: Visual explanations from deep networks via gradient-based localization. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV); 2017. p. 618–626. https://doi.org/10.1109/ICCV.2017.74.

[16] Nasir IM, Tehsin S, Damaševičius R, Maskeliūnas R. Integrating explanations into CNNs by adopting spiking attention block for skin cancer detection. Algorithms 2024;17:557. https://doi.org/10.3390/a17120557.

[17] Pramila RP, Subhashini R. Multi-center validation of ladybug beetle optimized convolutional capsule neural networks with explainable AI for skin cancer classification using dermography images. Afr J Biomed Res 2024;27:972–989. https://doi.org/10.53555/AJBR.v27i3.3246.