State identification of Parkinson's disease based on transfer learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Local field potential (LFP) signals are crucial for understanding deep brain stimulation (DBS) mechanisms and developing adaptive DBS for Parkinson's disease (PD) motor symptoms.
• Accurate identification of PD states from LFP signals is essential for effective treatment and research.

Purpose of the Study:

• To propose a Parkinson's disease state identification algorithm utilizing transfer learning for feature extraction.
• To develop an automated method for distinguishing pathological states in PD patients using LFP signal analysis.

Main Methods:

• Employed continuous wavelet transform (CWT) to convert 1D LFP signals into 2D gray-scalogram and color images.
• Designed a Bayesian-optimized random forest (RF) classifier integrated into the VGG16 model for image classification.
• Utilized gray-scalogram images for superior performance in PD state classification.

Main Results:

• The proposed algorithm achieved high accuracy (97.76%), precision (99.01%), recall (96.47%), and F1-score (97.73%).
• Gray-scalogram images demonstrated superior performance compared to color images.
• The algorithm outperformed established feature extractors like VGG19, InceptionV3, ResNet50, and MobileNet.

Conclusions:

• The developed algorithm accurately identifies PD patient states without manual feature extraction.
• This automated approach effectively assists clinicians in diagnosing and managing Parkinson's disease.
• The findings highlight the potential of LFP signal analysis combined with machine learning for PD management.