Efficient quantification of Parkinson's disease severity using augmented time-series data.

Area of Science:

• Biomedical Engineering
• Neurology
• Data Science

Background:

• Parkinson's disease diagnosis is subjective and physician-dependent, leading to variability.
• Objective and efficient diagnostic methods are crucial for timely and accurate Parkinson's disease detection.

Purpose of the Study:

• To develop an objective diagnostic method for Parkinson's disease using time-series sensor data.
• To evaluate the effectiveness of data augmentation techniques in improving machine learning model performance for Parkinson's disease classification.

Main Methods:

• Utilized PhysioNet dataset with vertical ground reaction forces from 93 Parkinson's patients and 73 healthy individuals.
• Applied data preprocessing and various data augmentation techniques (jittering, scaling, rotation, etc.).
• Evaluated models using one-dimensional convolutional neural networks (1D-ConvNet) and one-dimensional Transformer networks with 10-fold cross-validation.

Main Results:

• The best data augmentation strategy achieved 90.8% accuracy, 92.0% precision, 91.0% recall, and 91.0% F1 score.
• Significant improvements in classification performance were observed after applying data augmentation.
• Demonstrated enhanced model generalization and diagnostic reliability.

Conclusions:

• Data augmentation is vital for improving machine learning model performance on time-series sensor data for Parkinson's disease diagnosis.
• Selected data augmentation techniques enhance diagnostic reliability and model generalization.
• Offers insights for researchers utilizing sensor data in medical diagnostics.