Multiple Empirical Kernel Mapping Based Broad Learning System for Classification of Parkinson's Disease With...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Transcranial sonography (TCS) is increasingly utilized for Parkinson's disease (PD) diagnosis.
• Computer-aided diagnosis (CAD) systems for PD using TCS require robust classification components.
• Existing Broad Learning System (BLS) methods for classification have limitations in feature interpretability and representation due to simple random mapping.

Purpose of the Study:

• To develop an improved BLS algorithm for enhanced feature representation and interpretability in classification tasks.
• To enhance the performance of TCS-based computer-aided diagnosis for Parkinson's disease.

Main Methods:

• Proposed a novel Multiple Empirical Kernel Mapping based Broad Learning System (MEKM-BLS) algorithm.
• Implemented MEKM to map feature nodes to enhancement nodes, creating a more meaningful enhancement layer.
• Evaluated the MEKM-BLS algorithm using TCS data for Parkinson's disease diagnosis.

Main Results:

• The proposed MEKM-BLS algorithm demonstrated superior performance compared to the original BLS.
• MEKM-BLS achieved enhanced feature representation and interpretability within the feedforward neural network.
• Experimental results confirmed the effectiveness of MEKM-BLS in the context of PD diagnosis using TCS.

Conclusions:

• MEKM-BLS offers a significant advancement over standard BLS for classification tasks.
• The developed algorithm shows promise for improving the accuracy and reliability of Parkinson's disease diagnosis via TCS-based CAD systems.
• This approach contributes to the development of more interpretable and effective machine learning models in medical diagnostics.