Machine Learning for Neuroimaging with Scikit-Learn

TL;DR

This paper discusses how scikit-learn, a Python library, facilitates various machine learning techniques for analyzing high-dimensional neuroimaging data, enabling insights into brain function and structure.

Contribution

It demonstrates the application of scikit-learn's diverse algorithms to neuroimaging data analysis, highlighting its versatility and utility in the field.

Findings

Effective modeling of high-dimensional neuroimaging datasets

Use of supervised learning for decoding brain-behavior relationships

Unsupervised methods reveal hidden brain structures

Abstract

Statistical machine learning methods are increasingly used for neuroimaging data analysis. Their main virtue is their ability to model high-dimensional datasets, e.g. multivariate analysis of activation images or resting-state time series. Supervised learning is typically used in decoding or encoding settings to relate brain images to behavioral or clinical observations, while unsupervised learning can uncover hidden structures in sets of images (e.g. resting state functional MRI) or find sub-populations in large cohorts. By considering different functional neuroimaging applications, we illustrate how scikit-learn, a Python machine learning library, can be used to perform some key analysis steps. Scikit-learn contains a very large set of statistical learning algorithms, both supervised and unsupervised, and its application to neuroimaging data provides a versatile tool to study the brain.