Optimized MobileNetV3: a deep learning-based Parkinson's disease classification using fused images

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a progressive neurological disorder with diverse motor and non-motor symptoms, often including early vocal difficulties.
• Accurate and timely diagnosis of PD is crucial for effective patient management and improved quality of life.
• Current machine learning methods for PD detection often suffer from poor performance due to manual feature extraction limitations.

Purpose of the Study:

• To develop a robust deep learning model for the early and accurate diagnosis of Parkinson's disease.
• To enhance the classification performance of PD detection by optimizing a deep learning architecture.
• To address the limitations of traditional machine learning approaches in PD diagnosis.

Main Methods:

• A deep learning model based on MobileNetV3 was proposed for Parkinson's disease classification.
• The MobileNetV3 model was optimized using the Improved Dwarf Mongoose Optimization algorithm (IDMO) to boost performance.
• A Pyramid channel-based feature attention network (PCFAN) was employed to select critical features for enhanced classification.

Main Results:

• The proposed optimized MobileNetV3 model, incorporating PCFAN, demonstrated superior performance in PD detection.
• The approach achieved high accuracy (99.34%), sensitivity (98.53%), specificity (97.78%), and F-score (99.12%) on the PPMI and NTUA datasets.
• These results significantly outperform previous methods in identifying Parkinson's disease.

Conclusions:

• The developed deep learning framework offers a powerful tool for the early diagnosis of Parkinson's disease.
• The integration of MobileNetV3, IDMO, and PCFAN provides a highly accurate and efficient method for PD detection.
• This approach holds significant potential for improving clinical diagnostic tools for Parkinson's disease.