ParkEnNET: a majority voting-based ensemble transfer learning framework for early Parkinson's disease detection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Parkinson's Disease (PD) is a progressive neurodegenerative disorder impacting millions, particularly the elderly.
• Early diagnosis is crucial but challenging due to disease complexity and data limitations.
• Traditional deep learning models struggle with small, noisy medical datasets.

Purpose of the Study:

• To develop and validate ParkEnNET, a novel framework for early Parkinson's Disease detection.
• To leverage transfer learning and ensemble methods to overcome limitations of traditional deep learning models in medical image analysis.
• To improve diagnostic accuracy and efficiency for Parkinson's Disease using limited MRI data.

Main Methods:

• Developed ParkEnNET, a Majority Voting-Based Ensemble Transfer Learning Framework.
• Utilized pretrained deep learning models (VGGNet, ResNet-50, EfficientNet) for feature extraction from MRI data.
• Employed a majority voting ensemble strategy to integrate multiple model predictions, enhancing robustness.

Main Results:

• ParkEnNET achieved a diagnostic accuracy of 98.23% on Parkinson's Disease detection.
• Demonstrated high performance metrics: 100.0% precision, 95.24% recall, and 97.44% F1-score.
• Outperformed individual deep learning models, showcasing the effectiveness of the ensemble approach.

Conclusions:

• ParkEnNET offers a robust and accurate solution for early Parkinson's Disease detection, especially with limited datasets.
• The framework's strong performance on independent clinical data suggests significant potential for real-world clinical application.
• Enhanced early detection capabilities can lead to more timely interventions and improved patient outcomes for Parkinson's Disease.