An enhanced framework for Parkinson's disease severity prediction using improved optimization in multi-scale TCN.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Parkinson's Disease (PD) severity prediction is crucial for effective treatment and management.
• Conventional models struggle with complex PD data patterns and feature analysis.
• Deep learning offers potential but faces challenges in interpretability and parameter tuning.

Purpose of the Study:

• To develop an accurate and robust deep learning framework for predicting Parkinson's Disease severity.
• To address limitations of conventional models in analyzing complex PD data and features.
• To enhance clinical interpretability in PD severity prediction models.

Main Methods:

• An automated framework utilizing ensemble feature extraction and an adaptive multi-scale deep learning model (AMTCN).
• Feature extraction includes optimal weighted features (via Improved Archimedes Optimization Algorithm - IAOA), RBM features, and t-SNE features.
• IAOA optimizes feature selection and tunes AMTCN parameters for improved convergence and interpretability.

Main Results:

• The developed model achieved 94% accuracy and superior specificity compared to conventional methods.
• The framework offers clinical interpretability through IAOA-generated optimal weighted features, moving beyond black-box models.
• Improved performance leads to reduced diagnostic error rates for more precise PD evaluation.

Conclusions:

• The adaptive multi-scale deep learning model provides a highly accurate and interpretable solution for Parkinson's Disease severity prediction.
• This approach enhances diagnostic precision, aiding clinicians in treatment planning and disease management.
• The study demonstrates the efficacy of integrating advanced AI techniques for complex neurological disorder assessment.