The Cognitive Computing effort is a collaboration between Argonne’s Data Science and Learning division, University of California San Francisco, and Lawrence Berkeley and Lawrence Livermore national laboratories. Argonne has been applying machine and deep learning techniques to build predictive models for diagnosis of traumatic brain injury symptoms, prognosis of physical and neurological outcomes, and enhancement of medical images.

The goals of this project include advancing precision medicine with computational resources and methods, as well as advancing AI techniques for complex multi-modal data, predictive modeling, image enhancement, and low-data learning. The team has achieved highly accurate (greater than 85% AUC) predictive accuracy for abstract outcomes including returning to work, increased depression, as well as memory/language measures. Argonne has also significantly accelerated (from up to 24 hours to less than 30 minutes) a pipeline for generating macro-level structural connectomes via containerization, parallelization, and deep learning for anatomical segmentation of the brain. Importantly, this AI-based segmentation can produce annotations even for lower-quality CT images, the most common and quickest imaging modality for traumatic brain injury patients.