Integrating Explainability into RNNs for Amyotrophic Lateral Sclerosis Diagnosis

Such advancement of deep learning models, especially in the recurrent neural networks in the medical diagnostics had been very potent to predict the diagnoses of several diseases, including the Amyotrophic Lateral Sclerosis (ALS), though such dark-box nature of RNN also hampers in terms of its model interpretability, that is indispensable for the clinical area. A proposed study is the integration of newly devised explainability techniques, such as attention mechanisms, SHAP (SHapley Additive exPlanations), and Layer-wise Relevance Propagation (LRP), with RNNs to predict ALS diagnosis. On the dataset of 1,500 patient records suffering from ALS, it was observed that the model increased its transparency to 25%, with 85% accuracy in prediction and 90% user satisfaction in the clinical scenario. This conclusion opens the way for the potential of explainable RNNs in the improvement of ALS diagnosis and decision-making.