Research on a New AI Diagnostic Model with Strong Universality Based on Multilayer Perceptron Neural Networks

Parkinson’s Disease (PD) is a prevalent neurodegenerative disorder, and early prediction and diagnosis are crucial for managing the condition in patients. Vocal cord damage, an early symptom found in 90% of PD patients, suggests that acoustic features extracted from speech signals, combined with neural network models, could facilitate early detection of PD. Addressing the issue of poor feature universality in existing AI models for PD diagnosis through speech, this paper proposes a novel AI diagnostic model based on Multilayer Perceptron (MLP) neural networks. Fifteen acoustic features with strong universality were selected based on medical theory, and the model was trained using multiple datasets from different devices. Validation of the model showed a diagnostic accuracy of approximately 80% across various datasets (such as the Oxford Parkinson’s Disease Detection Dataset and MDVR-KCL), enhancing the generalizability and universality of AI models for diagnosing Parkinson’s Disease through speech analysis.