Parkinson's Disease Detection from Voice Recordings Using Associative Memories.

Area of Science:

• Neurology and Machine Learning
• Biomedical Signal Processing

Background:

• Parkinson's disease (PD) diagnosis is challenging due to its chronic and progressive nature.
• Early PD diagnosis is crucial for managing the condition and improving patient outcomes.
• Existing associative memory (AM) models for PD diagnosis lack mechanisms for irrelevant feature removal, potentially limiting classification performance.

Purpose of the Study:

• To enhance the smallest normalized difference associative memory (SNDAM) algorithm for improved PD classification.
• To introduce a learning reinforcement phase into the SNDAM algorithm, creating the ISNDAM model.
• To evaluate the classification performance of the proposed ISNDAM model using voice samples from PD patients and healthy individuals.

Main Methods:

• The study proposed an improved associative memory algorithm, ISNDAM, incorporating a learning reinforcement phase.
• Two publicly available datasets from the UCI Machine Learning Repository, comprising voice samples of early-stage PD patients and healthy individuals, were utilized.
• The ISNDAM model's efficiency was compared against seventy other models in the WEKA workbench and previous studies, with statistical significance analysis performed.

Main Results:

• ISNDAM achieved a classification accuracy of 99.48% on Dataset 1 and 99.66% on Dataset 2.
• The performance of ISNDAM surpassed other well-known algorithms, including ANN Levenberg-Marquardt (95.89%) and SVM RBF kernel (88.21%) on Dataset 1.
• Statistical significance tests confirmed that ISNDAM's classification performance is equivalent to models from previous studies.

Conclusions:

• The proposed ISNDAM algorithm effectively enhances classification performance for early Parkinson's disease diagnosis using voice samples.
• ISNDAM demonstrates superior accuracy and competitive performance compared to existing algorithms and previous research.
• The integration of a learning reinforcement phase is a key factor in ISNDAM's improved diagnostic capabilities.