Foundation models for discovering robust biomarkers of neurological disorders from dynamic functional connectivity

This work extends our BHI paper, where we proposed quantitative static FC metrics to evaluate the robustness of salient features identified by deep learning models. Code for our BHI work can be found here.

In this work, we further investigate the use of foundation models to extract dynamic functional connectivity (dFC) biomarkers.

This repository contains the implementation of the following metrics:-

1. Input Stability
2. Model Parameter Randomization Check
3. Label Randomization Check
4. Hub Assortativity Coefficient
5. Fidelity+
6. Hub Stability Index
7. Spatio-temporal Fidelity+

We package all these metrics in the RE-CONFIRM framework.

More updates to follow.