An XAI-Based Auditing Framework for Detecting Shortcut Learning in Visual Cardiac Murmur Screening

• Juliano Costa Reis UFG
• Ronaldo Martins da Costa UFG

DOI: https://doi.org/10.5753/eniac.2026.30854

Resumo

We propose an interpretability auditing framework for spectrogrambased cardiac murmur screening, combining Leave-One-Group-Out cross-validation, Grad-CAM, the sanity check of Adebayo et al., Deletion/Insertion fidelity metrics and a structured random-saliency control. Across 36 model instances on PhysioNet/CinC 2016, in-distribution AUC of 0.945 falls to 0.497 at an unseen site (36 of 36 instances), with severe overconfidence. Yet explanations remain faithful: sanity holds in 76.3% of maps and Grad-CAM beats random saliency on both fidelity metrics in all configurations. We further show that the reported disagreement between Deletion and Insertion is largely driven by the model’s operating point and by measurement choices. Faithful explanations do not imply a useful model.

Referências

Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Hardt, M., and Kim, B. (2018). Sanity checks for saliency maps. In Advances in Neural Information Processing Systems (NeurIPS), pages 9525–9536, Montreal, QC, Canada.

Alkhodari, M. and Fraiwan, L. (2021). Convolutional and recurrent neural networks for the detection of valvular heart diseases in phonocardiogram recordings. Computer Methods and Programs in Biomedicine, 200:105940.

Alrabie, S. and Barnawi, A. (2025). Are artificial intelligence models listening like cardiologists? Bridging the gap between artificial intelligence and clinical reasoning in heart-sound classification using explainable artificial intelligence. Bioengineering, 12(6):558.

Chattopadhay, A., Sarkar, A., Howlader, P., and Balasubramanian, V. N. (2018). GradCAM++: Generalized gradient-based visual explanations for deep convolutional networks. In 2018 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 839–847, Lake Tahoe, NV, USA.

Clifford, G. D. et al. (2016). Classification of normal/abnormal heart sound recordings: The PhysioNet/CinC challenge 2016. In 2016 Computing in Cardiology Conference (CinC), pages 609–612, Vancouver, BC, Canada.

DeGrave, A. J., Janizek, J. D., and Lee, S.-I. (2021). AI for radiographic COVID-19 detection selects shortcuts over signal. Nature Machine Intelligence, 3:610–619.

Geirhos, R. et al. (2020). Shortcut learning in deep neural networks. Nature Machine Intelligence, 2(11):665–673.

Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (ICML), PMLR 70, pages 1321–1330, Sydney, Australia.

Howard, A. et al. (2019). Searching for MobileNetV3. In 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pages 1314–1324, Seoul, Korea (South).

Kim, J., Park, G., and Suh, B. (2022). Classification of phonocardiogram recordings using vision transformer architecture. In 2022 Computing in Cardiology (CinC), volume 49, pages 1–4, Tampere, Finland.

Kim, S. et al. (2024). Artificial intelligence (AI) applications for point of care ultrasound (POCUS) in low-resource settings: A scoping review. Diagnostics, 14(15):1669.

Krishna, S. et al. (2024). The disagreement problem in explainable machine learning: A practitioner’s perspective. Transactions on Machine Learning Research.

Krones, F. et al. (2024). From theoretical models to practical deployment: A perspective and case study of opportunities and challenges in AI-driven cardiac auscultation research for low-income settings. PLOS Digital Health, 3(12):e0000437.

Li, X., Ng, G. A., and Schlindwein, F. S. (2022). Transfer learning in heart sound classification using Mel spectrogram. In 2022 Computing in Cardiology (CinC), volume 49, pages 1–4, Tampere, Finland.

Lundberg, S. M. and Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems (NeurIPS), pages 4765–4774, Long Beach, CA, USA.

Ma, S., Chen, J., and Ho, J. W. K. (2024). An edge-device-compatible algorithm for valvular heart diseases screening using phonocardiogram signals with a lightweight convolutional neural network and self-supervised learning. Computer Methods and Programs in Biomedicine, 243:107906.

Mehta, S. and Rastegari, M. (2022). MobileViT: Light-weight, general-purpose, and mobile-friendly vision transformer. In 10th International Conference on Learning Representations (ICLR).

Oliveira, J. et al. (2022). The CirCor DigiScope dataset: From murmur detection to murmur classification. IEEE Journal of Biomedical and Health Informatics, 26(6):2524–2535.

Özcan, F. (2024). Rapid detection and interpretation of heart murmurs using phonocardiograms, transfer learning and explainable artificial intelligence. Health Information Science and Systems, 12(1):43.

Petsiuk, V., Das, A., and Saenko, K. (2018). RISE: Randomized input sampling for explanation of black-box models. In Proceedings of the 29th British Machine Vision Conference (BMVC), paper 151, Newcastle, UK.

Ribeiro, M. T., Singh, S., and Guestrin, C. (2016). “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 (KDD), pages 1135–1144, San Francisco, CA, USA.

Roth, G. A. et al. (2020). Global burden of cardiovascular diseases and risk factors, 1990–2019: Update from the GBD 2019 study. Journal of the American College of Cardiology, 76(25):2982–3021.

Selvaraju, R. R. et al. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. In 2017 IEEE International Conference on Computer Vision (ICCV), pages 618–626, Venice, Italy.

Sundararajan, M., Taly, A., and Yan, Q. (2017). Axiomatic attribution for deep networks. In Proceedings of the 34th International Conference on Machine Learning (ICML), PMLR 70, pages 3319–3328, Sydney, Australia.

Tan, M. and Le, Q. V. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. In Proceedings of the 36th International Conference on Machine Learning (ICML), PMLR 97, pages 6105–6114, Long Beach, CA, USA.

Wang, H. et al. (2020). Score-CAM: Score-weighted visual explanations for convolutional neural networks. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 24–25, Seattle, WA, USA.

Zech, J. R., Badgeley, M. A., Liu, M., Costa, A. B., Titano, J. J., and Oermann, E. K. (2018). Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLOS Medicine, 15(11):e1002683.