Explainable foundation model framework for Autism characterization from EEG signals.

Mentors

• 

  Archana Venkataraman

  Associate Professor (ECE, BME)

• 

  Deeksha Shama

  PhD Candidate

Project Description

Autism spectrum disorder (ASD) is challenging to diagnose and characterize due to its heterogeneous behavioral presentation and complex neurobiological underpinnings. Electroencephalography (EEG) has shown promise for ASD characterization by capturing atypical neural oscillations and sensory processing dynamics that may serve as objective biomarkers of underlying brain function. However, traditional EEG analysis methods that rely on manual feature extraction often suffer from poor generalizability, limited scalability, and sensitivity to noise and preprocessing choices. Alternatively, deep learning approaches offer powerful modeling capabilities for ASD classification; however they are often treated as black-box models with limited interpretability, reducing their reliability and clinical trustworthiness in healthcare settings. In this project, we will compare pretrained foundation models for classifying ASD from EEG signals and perform explainability analyses to identify discriminative features across temporal, spectral, and spatial domains at the subject, group, and longitudinal cohort levels.

Timeline

Weeks 1-2: Literature review and familiarizing with the data.
Weeks 3-5: Implementing foundation model pipeline for autism classification from EEG.
Weeks 6-10: Longitudinal analyses with relevant EEG features and behavioral phenotypes; Documentation and presentation.