LATENT SUBGROUP IDENTIFICATION IN IMAGE-ON-SCALAR REGRESSION

Area of Science:

• Neuroimaging
• Statistical Modeling
• Computational Neuroscience

Background:

• Image-on-scalar regression is common for analyzing brain activity and scalar characteristics.
• Neuroimaging studies, like the Adolescent Brain Cognitive Development (ABCD) Study, reveal heterogeneous associations across individuals.
• Existing methods struggle to identify subgroups with homogeneous within-group associations and heterogeneous across-group associations.

Purpose of the Study:

• To propose a novel latent subgroup image-on-scalar regression model (LASIR).
• To effectively analyze large-scale, multisite neuroimaging data with diverse sociodemographics.
• To identify population subgroups with distinct brain activity-clinical measure associations.

Main Methods:

• Developed the Latent Subgroup Image-on-Scalar Regression (LASIR) model.
• Incorporated latent subgroups and group-specific, spatially varying effects.
• Utilized an efficient stochastic expectation maximization algorithm for inference.

Main Results:

• LASIR successfully identifies subgroups with homogeneous associations within and heterogeneous associations across groups.
• Demonstrated superior performance of LASIR compared to existing methods in simulations.
• Applied LASIR to the Adolescent Brain Cognitive Development (ABCD) study data.

Conclusions:

• LASIR provides an effective approach for subgroup identification in neuroimaging data.
• The model leverages individual characteristics for group allocation.
• Reproducible code is available on Github for public use.