SCoRS--A Method Based on Stability for Feature Selection and Mapping inNeuroimaging [corrected].

Area of Science:

• Neuroimaging
• Machine Learning
• Biostatistics

Background:

• Feature selection (FS) is crucial for neuroimaging classification, improving accuracy and interpretability.
• Current FS methods often prioritize accuracy over reliable feature set interpretation.
• A need exists for FS techniques that yield stable and interpretable feature sets in neuroimaging.

Purpose of the Study:

• To propose a novel feature selection method, SCoRS (survival count on random subsamples), for neuroimaging data.
• To enhance the reliability and interpretability of feature selection in brain imaging classification tasks.
• To evaluate SCoRS's performance in classifying depressed patients versus healthy individuals using fMRI data.

Main Methods:

• SCoRS is based on Stability Selection theory, selecting features that are stable across data perturbations.
• Data perturbation is achieved through iterative sub-sampling of both features (subspaces) and examples.
• The method was applied to functional magnetic resonance imaging (fMRI) data from depressed patients and healthy controls.

Main Results:

• The proposed SCoRS method demonstrates potential in identifying relevant and stable features for classification.
• Application to fMRI data for depression classification showed promising results in feature identification.
• The method facilitates more reliable interpretation by focusing on stable feature subsets.

Conclusions:

• SCoRS offers a robust approach to feature selection in neuroimaging, balancing accuracy and interpretability.
• The stability-based selection ensures more reliable identification of discriminative brain features.
• This method holds promise for clinical applications, such as diagnosing depression using neuroimaging.