Enhanced Parkinson's Detection Through Multisyllabic Vowel and Spontaneous Dialogue Analysis Using Graph Convolutional Networks

Long-term, progressive, and degenerative, Parkinson's disease (PD) disturbs the brain and spinal cord. Typically, the symptoms gradually manifest, beginning with motor system symptoms and possibly progressing to non-motor symptoms as the disease advances. Nearly all persons with PD will eventually experience difficulties with their speech. Despite the fact that there have been several computer-based approaches using patients' speech for early PD diagnosis, these methods have primarily focused on a predefined articulation test, such as analyzing individuals' monosyllabic pronunciation to identify if they may have PD. Additionally, a complete feature representation cannot be provided by relying just on conventional speech analysis approaches to extract features from a single view of the audio. Five monosyllabic vowels and an impromptu conversation are among the assessments of PD patients' pronunciation that this research examines. For the purpose of detecting PD in these two test sets, a convolutional deep learning network that is graph-based is developed and suggested. To begin, in order to prepare the data for processing, multisource data extracts various sets of speech features. Running City Game Optimizer (RCGO) is then used to identify the most relevant signal features, which enhances the classification accuracy. First, to learn a network for the corpus and then to construct a single graph for the corpus using co-occurrence relations. Using the existing class labels as supervision, the suggested model learns the embeddings for both normal and PD signals simultaneously after being initialized with a one-hot representation for PD classification. Both sets of testing achieved the best PD identification rates, which is a benchmark for future academic work in the area and beats out a plethora of well-known pattern recognition systems around the world.