Particle swarm optimization framework for Parkinson's disease prediction.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Artificial Intelligence in Medicine

Background:

• Early diagnosis of Parkinson's disease (PD) is hindered by subtle initial symptoms, necessitating advanced detection methods.
• Vocal biomarkers offer a non-invasive avenue for PD assessment, but their diagnostic potential requires sophisticated analytical frameworks.
• Current diagnostic approaches may not fully capture the complexity of early PD indicators, leading to delayed intervention.

Purpose of the Study:

• To develop and evaluate an advanced machine learning framework integrating particle swarm optimization (PSO) for enhanced Parkinson's disease detection using vocal biomarkers.
• To unify acoustic feature selection and classifier hyperparameter tuning within a single computational architecture for improved predictive accuracy.
• To assess the practical viability and clinical implications of PSO-optimized decision support systems for early neurodegenerative disease detection.

Main Methods:

• A novel machine learning framework leveraging particle swarm optimization (PSO) was developed to optimize both feature selection and classifier hyperparameters for PD detection.
• The PSO-enhanced models were systematically evaluated on two distinct clinical datasets (Dataset 1: 1,195 records, 24 features; Dataset 2: 2,105 records, 33 multidimensional features).
• Performance metrics including accuracy, sensitivity, specificity, and Area Under the Curve (AUC) were compared against traditional machine learning classifiers.

Main Results:

• For Dataset 1, the PSO model achieved 96.7% testing accuracy, outperforming the best traditional classifier (Bagging) by 2.6%, with 99.0% sensitivity and 94.6% specificity.
• Dataset 2 demonstrated even greater improvements, with the PSO model reaching 98.9% accuracy (3.9% higher than LGBM) and a near-perfect AUC of 0.999.
• The PSO optimization demonstrated practical viability with an average training time of 250.93 seconds for Dataset 2, indicating reasonable computational overhead.

Conclusions:

• The proposed PSO-enhanced machine learning framework significantly improves the accuracy and discriminative capability for early Parkinson's disease detection using vocal biomarkers.
• Intelligent optimization techniques like PSO hold substantial potential for developing practical clinical decision support systems for neurodegenerative diseases.
• These findings suggest a promising direction for advancing early diagnosis and intervention strategies in Parkinson's disease management.