Brain Biomarker Interpretation in ASD Using Deep Learning and fMRI

Xiaoxiao Li, Nicha C. Dvornek, Juntang Zhuang, Pamela Ventola, James S. Duncan

Abstract

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder. Finding the biomarkers associated with ASD is extremely helpful to understand the underlying roots of the disorder and can lead to earlier diagnosis and more targeted treatment. Although Deep Neural Networks (DNNs) have been applied in functional magnetic resonance imaging (fMRI) to identify ASD, understanding the data-driven computational decision making procedure has not been previously explored. Therefore, in this work, we address the problem of interpreting reliable biomarkers associated with identifying ASD; specifically, we propose a 2-stage method that classifies ASD and control subjects using fMRI images and interprets the saliency features activated by the classifier. First, we trained an accurate DNN classifier. Then, for detecting the biomarkers, different from the DNN visualization works in computer vision, we take advantage of the anatomical structure of brain fMRI and develop a frequency-normalized sampling method to corrupt images. Furthermore, in the ASD vs. control subjects classification scenario, we provide a new approach to detect and characterize important brain features into three categories. The biomarkers we found by the proposed method are robust and consistent with previous findings in the literature. We also validate the detected biomarkers by neurological function decoding and comparing with the DNN activation maps.

Introduction

While Deep Neural Networks (DNNs) have shown great promise in identifying Autism Spectrum Disorder (ASD) using functional Magnetic Resonance Imaging (fMRI), they are notoriously "black boxes." For a clinician, a high accuracy score is insufficient; they need to understand which brain regions are abnormal to provide targeted treatment and early diagnosis. Existing visualization techniques often produce blurry, imprecise results that are difficult to map back to specific anatomical structures. Furthermore, fMRI datasets are often small and imbalanced, leading to biased interpretations. This study bridges the gap between high-performance deep learning and interpretable clinical biomarkers.

Methodology

The researchers propose a two-stage pipeline to classify ASD and interpret the underlying decision-making process.

Overview of the proposed two-stage pipeline for ASD classification and biomarker interpretation.

1. Classification

A 3D Convolutional Neural Network (specifically a "2CC3D" model) is trained on 4D fMRI data. To handle the temporal aspect, 4D sequences are converted into 2-channel 3D images representing the mean and standard deviation of voxel intensity over a sliding window.

2. Interpretation via Corrupting Strategy

To identify important biomarkers, the method systematically "corrupts" specific Regions of Interest (ROIs) based on the AAL brain atlas.

  • Frequency-Normalized Sampling: Instead of simple occlusion, an ROI is replaced with samples from other subjects. A weighting system based on Pearson correlation ensures the corrupted data is representative and prevents bias from class imbalances.
  • Statistical Validation: Corrupted images are passed back through the DNN. Significant changes in prediction (measured via Jensen-Shannon Divergence and Wilcoxon signed-rank tests) flag an ROI as a biomarker.

Results and Discussion

The study utilized three datasets: a synthetic dataset for proof-of-concept, the Biopoint dataset (task-based fMRI), and ABIDE I (resting-state fMRI).

Performance and Biomarker Identification

The 2CC3D model achieved high classification accuracy: 87.1% on Biopoint and 85.3% on ABIDE. Beyond accuracy, the interpretation method successfully categorized ROIs into three groups: those important for identifying ASD, those for controls, and those significant for both.

Identified brain biomarkers in the Biopoint dataset, categorized by their importance to ASD and control groups.

The identified biomarkers—including regions related to the "default mode network," functional connectivity, and motor processing—align closely with existing neurological literature. Consistency across both task-based and resting-state fMRI suggests these are robust indicators of ASD.

Conclusion

This work introduces a robust framework for interpreting DNN decisions in medical imaging. By leveraging anatomical structures and a novel frequency-normalized sampling strategy, the method provides clinically relevant biomarkers. Future work may involve using generative models like GANs for more precise image corruption and further validation on larger multi-site datasets to ensure generalizability.

Summary updated

Abstract

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder. Finding the biomarkers associated with ASD is extremely helpful to understand the underlying roots of the disorder and can lead to earlier diagnosis and more targeted treatment. Although Deep Neural Networks (DNNs) have been applied in functional magnetic resonance imaging (fMRI) to identify ASD, understanding the data-driven computational decision making procedure has not been previously explored. Therefore, in this work, we address the problem of interpreting reliable biomarkers associated with identifying ASD; specifically, we propose a 2-stage method that classifies ASD and control subjects using fMRI images and interprets the saliency features activated by the classifier. First, we trained an accurate DNN classifier. Then, for detecting the biomarkers, different from the DNN visualization works in computer vision, we take advantage of the anatomical structure of brain fMRI and develop a frequency-normalized sampling method to corrupt images. Furthermore, in the ASD vs. control subjects classification scenario, we provide a new approach to detect and characterize important brain features into three categories. The biomarkers we found by the proposed method are robust and consistent with previous findings in the literature. We also validate the detected biomarkers by neurological function decoding and comparing with the DNN activation maps.
BibTeX
@inproceedings{li2018brain,
  author = {Xiaoxiao Li and Nicha C. Dvornek and Juntang Zhuang and Pamela Ventola and James S. Duncan},
  title = {{Brain Biomarker Interpretation in ASD Using Deep Learning and fMRI}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2018},
  doi = {10.1007/978-3-030-00931-1\_24},
}
Figure