Towards Trustworthy AI in Cardiology (AIME2024)

This repository contains supplementary material accompanying the paper entitled "Towards Trustworthy AI in Cardiology: A Comparative Analysis of Explainable AI Methods for Electrocardiogram Interpretation Towards Trustworthy AI in Cardiology" submitted to the 22nd International Conference on Artificial Intelligence in Medicine, Salt Lake City, Utah, USA, July 9-12 2024 (AIME2024). https://doi.org/10.1007/978-3-031-66535-6_36

Citation:

@InProceedings{Gumpfer2024,
  author    = {Gumpfer, Nils and Borislav Dinov and Samuel Sossalla and Michael Guckert and Jennifer Hannig},
  booktitle = {22nd International Conference on Artificial Intelligence in Medicine, AIME 2024, Salt Lake City, UT, USA, July 9 - 12, 2024, Proceedings},
  title     = {{Towards Trustworthy {AI} in Cardiology: A Comparative Analysis of Explainable {AI} Methods for Electrocardiogram Interpretation}},
  chapter   = {36},
  doi       = {10.1007/978-3-031-66535-6_36},
  editor    = {Finkelstein, Josef and Moskovitch, Robert and Parimbelli, Enea},
  pages     = {350--361},
  publisher = {Springer Nature Switzerland AG},
  series    = {Lecture Notes in Computer Science},
  volume    = {14845},
  month     = {07},
  year      = {2024},
}

Data

We used electrocardiograms from the PTB-XL database: https://physionet.org/content/ptb-xl/1.0.3/ To download the records used in the examples below and to setup the pip environment, run the prepare.sh script. For the experiments, we used Python 3.10.

Models

The readily trained models are provided in .h5 format for evaluation purposes.

Code

For the calculation of explanations, we used the sign-xai and shap python packages:

SIGN: https://github.com/nilsgumpfer/SIGN-XAI
SHAP: https://github.com/shap/shap

You can run the example_xai.py script to generate the explanations yourself.

Atrioventricular block

AVB_gradient ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient (Zurada et al., 1994) for predicted class atrioventricular block (prob. 97%).

AVB_gradient_x_input ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×Input (Shrikumar et al., 2017) for predicted class atrioventricular block (prob. 97%).

AVB_gradient_x_sign ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×SIGN (Gumpfer et al., 2023) for predicted class atrioventricular block (prob. 97%).

AVB_lrp_alpha_1_beta_0 ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-α₁β₀ (Bach et al., 2015) for predicted class atrioventricular block (prob. 97%).

AVB_lrp_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε with ε = 0.5 ⋅ σ(x) (Bach et al., 2015) for predicted class atrioventricular block (prob. 97%).

AVB_lrpsign_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε SIGN with ε = 0.5 ⋅ σ(x) and SIGN as input layer rule (Gumpfer et al., 2023) for predicted class atrioventricular block (prob. 97%).

AVB_deep_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method DeepSHAP (Lundberg et al., 2017) for predicted class atrioventricular block (prob. 97%).

AVB_grad_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method GradSHAP (Lundberg et al., 2017) for predicted class atrioventricular block (prob. 97%).

AVB_grad_cam_timeseries ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Grad-CAM (Selvaraju et al., 2020) for predicted class atrioventricular block (prob. 97%).

Myocardial ischemia

ISCH_gradient ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient (Zurada et al., 1994) for predicted class myocardial ischemia (prob. 69%).

ISCH_gradient_x_input ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×Input (Shrikumar et al., 2017) for predicted class myocardial ischemia (prob. 69%).

ISCH_gradient_x_sign ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×SIGN (Gumpfer et al., 2023) for predicted class myocardial ischemia (prob. 69%).

ISCH_lrp_alpha_1_beta_0 ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-α₁β₀ (Bach et al., 2015) for predicted class myocardial ischemia (prob. 69%).

ISCH_lrp_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε with ε = 0.5 ⋅ σ(x) (Bach et al., 2015) for predicted class myocardial ischemia (prob. 69%).

ISCH_lrpsign_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε SIGN with ε = 0.5 ⋅ σ(x) and SIGN as input layer rule (Gumpfer et al., 2023) for predicted class myocardial ischemia (prob. 69%).

ISCH_deep_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method DeepSHAP (Lundberg et al., 2017) for predicted class myocardial ischemia (prob. 69%).

ISCH_grad_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method GradSHAP (Lundberg et al., 2017) for predicted class myocardial ischemia (prob. 69%).

ISCH_grad_cam_timeseries ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Grad-CAM (Selvaraju et al., 2020) for predicted class myocardial ischemia (prob. 69%).

Right bundle branch block

RBBB_gradient ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient (Zurada et al., 1994) for predicted class right bundle branch block (prob. 99%).

RBBB_gradient_x_input ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×Input (Shrikumar et al., 2017) for predicted class right bundle branch block (prob. 99%).

RBBB_gradient_x_sign ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×SIGN (Gumpfer et al., 2023) for predicted class right bundle branch block (prob. 99%).

RBBB_lrp_alpha_1_beta_0 ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-α₁β₀ (Bach et al., 2015) for predicted class right bundle branch block (prob. 99%).

RBBB_lrp_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε with ε = 0.5 ⋅ σ(x) (Bach et al., 2015) for predicted class right bundle branch block (prob. 99%).

RBBB_lrpsign_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε SIGN with ε = 0.5 ⋅ σ(x) and SIGN as input layer rule (Gumpfer et al., 2023) for predicted class right bundle branch block (prob. 99%).

RBBB_deep_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method DeepSHAP (Lundberg et al., 2017) for predicted class right bundle branch block (prob. 99%).

RBBB_grad_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method GradSHAP (Lundberg et al., 2017) for predicted class right bundle branch block (prob. 99%).

RBBB_grad_cam_timeseries ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Grad-CAM (Selvaraju et al., 2020) for predicted class right bundle branch block (prob. 99%).

Left bundle branch block

LBBB_gradient ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient (Zurada et al., 1994) for predicted class left bundle branch block (prob. 98%).

LBBB_gradient_x_input ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×Input (Shrikumar et al., 2017) for predicted class left bundle branch block (prob. 98%).

LBBB_gradient_x_sign ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Gradient×SIGN (Gumpfer et al., 2023) for predicted class left bundle branch block (prob. 98%).

LBBB_lrp_alpha_1_beta_0 ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-α₁β₀ (Bach et al., 2015) for predicted class left bundle branch block (prob. 98%).

LBBB_lrp_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε with ε = 0.5 ⋅ σ(x) (Bach et al., 2015) for predicted class left bundle branch block (prob. 98%).

LBBB_lrpsign_epsilon_0_5_std_x ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method LRP-ε SIGN with ε = 0.5 ⋅ σ(x) and SIGN as input layer rule (Gumpfer et al., 2023) for predicted class left bundle branch block (prob. 98%).

LBBB_deep_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method DeepSHAP (Lundberg et al., 2017) for predicted class left bundle branch block (prob. 98%).

LBBB_grad_shap ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method GradSHAP (Lundberg et al., 2017) for predicted class left bundle branch block (prob. 98%).

LBBB_grad_cam_timeseries ECG subsample from PTB-XL (Wagner et al., 2020) with heatmap overlay (red) based on XAI method Grad-CAM (Selvaraju et al., 2020) for predicted class left bundle branch block (prob. 98%).

References

Bach, S., Binder, A., Montavon, G., Klauschen, F., Müller, K.R., Samek, W.: On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PLoS One 10(7), 1–46 (2015)

Gumpfer, N., Prim, J., Keller, T., Seeger, B., Guckert, M., Hannig, J.: SIGNed explanations: Unveiling relevant features by reducing bias. Information Fusion 99, 101883 (2023)

Lundberg, S.M., Lee, S.: A unified approach to interpreting model predictions. In: Guyon, I., von Luxburg, U., Bengio, S., Wallach, H.M., Fergus, R., Vishwanathan, S.V.N., Garnett, R. (eds.) Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA. pp. 4765–4774 (2017)

Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: GradCAM: Visual explanations from deep networks via gradient-based localization. Int. J. Comput. Vision 128(2), 336–359 (2020)

Shrikumar, A., Greenside, P., Kundaje, A.: Learning important features through propagating activation differences. In: Precup, D., Teh, Y.W. (eds.) Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017. Proceedings of Machine Learning Research, vol. 70, pp. 3145–3153. PMLR (2017)

Wagner, G.S., Macfarlane, P., Wellens, H., Josephson, M., Gorgels, A., Mirvis, D.M., Pahlm, O., Surawicz, B., Kligfield, P., Childers, R., Gettes, L.S.: AHA/ACCF/HRS recommendations for the standardization and interpretation of the electrocardiogram - part VI: Acute ischemia/infarction. Circulation 119(10), e262–e270 (2009)

Wagner, P., Strodthoff, N., Bousseljot, R.D., Kreiseler, D., Lunze, F.I., Samek, W., Schaeffter, T.: PTB-XL, a large publicly available electrocardiography dataset. Scientific Data 7(1), 154 (2020)

Zurada, J., Malinowski, A., Cloete, I.: Sensitivity analysis for minimization of input data dimension for feedforward neural network. In: Proceedings of IEEE International Symposium on Circuits and Systems - ISCAS ’94. vol. 6, pp. 447–450 vol.6 (1994)