Document Type



Understanding cognitive states from fMRI data have yet to be investigated to its full extent due to its complex nature. In this work, the problem of understanding cognitive fatigue among TBI patients has been formulated as a multi-class classification problem. We built a Spatio-temporal encoder model using convolutions and LSTMs as the building blocks to extract spatial features and to model the 4D nature of fMRI scans. To learn a better representation of the data and the condition, we used a self-supervised learning technique called "Contrastive Learning" to pretrain our encoder with a public dataset BOLD5000 and further fine-tuned our labeled dataset to predict cognitive fatigue. Furthermore, we present an fMRI dataset that contains scans from a mix of Traumatic Brain Injury (TBI) patients and healthy controls (HCs) while performing a series of standardized N-back cognitive tasks. This method establishes a state-of-the-art technique to analyze cognitive fatigue from fMRI data and beats previous approaches to solve this problem with different modalities. Besides, the ability of our models to take in raw fMRI scans (noisy images with artifacts output directly from the scanner) eliminates the need to implement a manual signal processing pipeline that varies based on the scanner used. Finally, we study the impact of different brain regions contributing to CF. The proposed technique outperforms the state-of-the-art method by over 13 percent on this dataset.


Computer Sciences | Physical Sciences and Mathematics

Publication Date





Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.