Speech signal-based accurate neurological disorders detection using convolutional neural network and recurrent neural network based deep network

Abstract

Neurological diseases often manifest in subtle alterations to the human voice due to damage in the sound-related regions of the brain. Leveraging advancements in artificial intelligence (AI) technologies, computers can discern minute variations in sound imperceptible to the human ear, enabling rapid and precise diagnostic support. This paper presents a novel approach to neurological disease classification utilizing voice recordings of individuals diagnosed with various neurological conditions alongside healthy controls. By employing AI techniques, particularly a hybrid deep network framework integrating Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), we aimed to classify one-sentence audio inputs of Multiple Sclerosis (MS) patients, healthy individuals, and other neurological diseases. In our dataset, we have compiled audio recordings from 95 healthy individuals, 99 individuals diagnosed with multiple sclerosis (MS), and 96 individuals with other neurological disorders. Of these, 20 % of the data was reserved for testing. Our proposed architecture achieved remarkable performance metrics in experimental evaluations, exhibiting 96.55 % accuracy, 98.25 % specificity, 96.49 % sensitivity, 96.97 % precision, and 96.56 % F1-Score. The results obtained are more successful compared to the methods of AlexNet from scratch, fine-tuned AlexNet, Long Short-Term Memory (LSTM) based CNN, and Gated Recurrent Unit (GRU) based CNN. The results of our study highlight the potential of this framework to be integrated into clinical diagnostic workflows, providing clinicians with an effective tool for early and precise detection of neurological diseases, ultimately improving patient outcomes through timely intervention and personalized treatment strategies.