Interpreting Models Interpreting Brain Dynamics

This repository provides code to replicate the findings in: Interpreting Models Interpreting Brain Dynamics by
M. M. Rahman, U. Mahmood, N. Lewis, H. Gazula, A. Fedorov, Z. Fu, V. D. Calhoun & S. M. Plis.

Overview

Brain dynamics are highly complex and yet hold the key to understanding brain function and dysfunction. The dynamics captured by resting-state functional magnetic resonance imaging data are noisy, high-dimensional, and not readily interpretable. The typical approach of reducing this data to low-dimensional features and focusing on the most predictive features comes with strong assumptions and can miss essential aspects of the underlying dynamics. In contrast, introspection of discriminatively trained deep learning models may uncover disorder-relevant elements of the signal at the level of individual time points and spatial locations. Yet, the difficulty of reliable training on high-dimensional low sample size datasets and the unclear relevance of the resulting predictive markers prevent the widespread use of deep learning in functional neuroimaging. In this work, we introduce a deep learning framework to learn from high-dimensional dynamical data while maintaining stable, ecologically valid interpretations. Results successfully demonstrate that the proposed framework enables learning the dynamics of resting-state fMRI directly from small data and capturing compact, stable interpretations of features predictive of function and dysfunction. A pre-print version can be accessed from here

Data

• Function Biomedical Informatics Research Network (FBIRN)
• Open Access Series of Imaging Studies (OASIS)
• Autism Brain Imaging Data Exchange (ABIDE)
• Human Connectome Project (HCP) (For pre-training only)

Experiments and Scripts

In the scripts folder, all of the scripts required to build and evaluate Standard Machine Learning (SML) models, pretrain whole MILC, generate post-hoc explanations evaluate explanations using RAR method are provided:

• run_sml.py: is used to build and evaluate standard machine learning models from the raw data (ICA time-courses).

• run_milc_pretraining.py: is used to pretrain whole MILC model leveraging only healthy (control) subjects from HCP.

• run_downstream_model: is used to train and evaluate downstream models for different disorders.

• run_downstream_sal_compute.py: is used to compute saliency maps/feature attributions per (model-disorder) pair.

• run_rar_fresh.py: is used to build new SML models using RAR and SVM with top (5% - 30%) features guided by computed Saliency.

• run_random_rar.py: is used to build SML models using random (5% - 30%)feature selection and SVM models.

• saliency_interp_n_vis.ipynb is used to visualize saliency maps, to analyze group-level behavior (Functional Network Connectivity,Timing Characteristics etc.).

How to Run:

• For building standard machine learning models on raw data:

python run_sml.py dataset_id

• For pretraining whole MILC using HCP data:

python run_milc_pretraining.py

• For building downstream whole MILC model using disorder specific data:

python run_downstream_model.py dataset_id

• For generating post-hoc explanations per (model-disorder) pair:

python run_downstream_sal_compute.py dataset_id

• For evaluation of posthoc explanations using RAR and SVM:

python run_rar_fresh.py dataset_id

dataset_id options:

• FBIRN: 0
• OASIS: 2
• ABIDE: 3