Detection of Parkinson's Disease from Voice Signals Using Explainable Artificial Intelligence

Neurodegenerative diseases such as Alzheimer's and Parkinson's disease have presented one of the major health challenges facing the world today, especially with the increasing prevalence seen with increasing aging populations. In this regard, AI has been emerged as a promising method for early detection and intervention. Therefore, in the paper, an explainable approach for deep learning in voice recordings toward diagnosing Parkinson's disease is presented. It uses a modified Attention ResNet18 architecture with Squeeze-and-Excitation blocks and utilizes Mel-Spectrogram features obtained from audio signals. Grad-CAM, Integrated Gradients, and LIME were used to provide the explanations generated by the model during the classification process. Experimental results presented an accuracy of 97.8<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> for classification, thus demonstrating the ability of Explainable AI to detect non-invasive PD.