Accelerating Hyperparameter Tuning in Machine Learning for Alzheimer's Disease With High Performance Computing.

Area of Science:

• Computational neuroscience
• Biomedical data science
• Artificial intelligence in medicine

Background:

• Massive datasets combining blood biomarkers and MRI necessitate advanced machine learning (ML) algorithms and hardware accelerators (GPUs, FPGAs).
• ML shows promise for early Alzheimer's disease (AD) diagnosis, but increasing algorithmic complexity, like hyperparameter tuning, raises computational demands.
• Accelerating high-performance ML for AD is a critical research challenge.

Purpose of the Study:

• To develop and evaluate a multicore high-performance Support Vector Machine (SVM) hyperparameter tuning workflow.
• To speed up ML-driven analysis for AD diagnosis.
• To demonstrate the workflow's effectiveness on public MRI data.

Main Methods:

• Implemented a multicore high-performance SVM hyperparameter tuning workflow.
• Utilized 100 times repeated 5-fold cross-validation for robust evaluation.
• Applied the model to public MRI data, incorporating demographic factors (age, sex, education) for AD prediction.

Main Results:

• Achieved a 96% increase in computational efficiency.
• Demonstrated the model's applicability to AD diagnosis using MRI and demographic data.
• The workflow is adaptable to other ML algorithms like random forest, logistic regression, and xgboost.

Conclusions:

• The developed high-performance hyperparameter tuning workflow significantly accelerates ML for AD.
• This advancement has implications for future diagnostic biomarker applications in AD.
• The methodology offers a scalable solution for complex ML tasks in neurodegenerative disease research.