oFVSD: a Python package of optimized forward variable selection decoder for high-dimensional neuroimaging data.

Area of Science:

• Neuroimaging
• Machine Learning
• Data Science

Background:

• High-dimensional neuroimaging data presents challenges for machine learning decoding due to the large feature-to-observation ratio.
• Conventional machine learning models struggle with optimizing feature selection in complex, high-dimensional datasets.

Purpose of the Study:

• To introduce an efficient and high-performance decoding package, the optimized forward variable selection decoder (oFVSD).
• To automate the identification of optimal feature subsets and hyperparameters for machine learning models in neuroimaging data analysis.

Main Methods:

• Implemented a forward variable selection (FVS) algorithm integrated with hyper-parameter optimization for 18 machine learning models.
• Utilized k-fold cross-validation to evaluate feature subsets and optimize model performance.
• Applied the oFVSD pipeline to 1,113 structural magnetic resonance imaging (MRI) datasets for sex classification and age regression.

Main Results:

• The oFVSD pipeline demonstrated superior performance compared to models without FVS and those using the Boruta algorithm.
• Achieved an average increase of approximately 0.20 in correlation coefficient for regression and 8% for classification tasks.
• Confirmed that parallel computation significantly reduced the processing time for high-dimensional MRI data.

Conclusions:

• The oFVSD toolbox effectively enhances the performance of both classification and regression machine learning models in neuroimaging.
• The open-source Python package offers a valuable solution for researchers aiming to improve decoding accuracy with high-dimensional data.
• oFVSD shows potential for application across various neuroimaging modalities beyond the demonstrated MRI use case.