A Convolutional Neural Network Model for Classifying Resting Tremor Amplitude in Parkinson's Disease

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Resting tremor (RT) is a primary Parkinson's Disease (PD) symptom.
• Current RT assessment relies on the subjective Unified Parkinson's Disease Rating Scale (UPDRS) 3.17.
• Subjectivity in UPDRS 3.17 introduces diagnostic biases.

Purpose of the Study:

• To evaluate a Convolutional Neural Network (CNN) for multiclass classification of RT amplitude in PD.
• To compare CNN performance against traditional machine learning models.
• To develop a more objective and accurate method for RT assessment.

Main Methods:

• Utilized a public dataset from 3-axis accelerometers on PD patients and healthy subjects.
• Extracted resting data during UPDRS assessments and applied automatic segmentation.
• Developed a 7-layer CNN and employed 5-fold cross-validation for model testing.

Main Results:

• The CNN model achieved an average accuracy of 95.94% in classifying RT amplitude.
• The CNN significantly outperformed Random Forest, SVM, and Decision Trees.
• Demonstrated superior accuracy in tremor classification compared to traditional methods.

Conclusions:

• CNNs can efficiently classify RT amplitude in PD patients.
• The proposed CNN model offers enhanced accuracy and objectivity in RT assessment.
• This AI-driven approach can simplify diagnostic processes for Parkinson's Disease.