Predicting recovery after TBI: Development and comparison of MR-supplemented models using non-parametric and machine learning multimodal fusion

Project Summary/Abstract (30 lines max) The ability to leverage early biomarkers, clinical, and demographic data to accurately predict a patient’s likely recovery trajectory following moderate-to-severe traumatic brain injury (TBI) is paramount to allow definition of appropriate therapeutic strategies and to evaluate medical decision-making in the context of critical decisions

such as early withdrawal of life supporting therapies. Prognostication of neurofunctional recovery following TBI, however, is known to be as critical as challenging. Extensive work has shown that current approaches suffer from high variability across physicians and medical centers, as well as a tendency for overestimation of poor

outcomes and underestimation of positive outcomes, and to be affected by non-clinical factors such as geographic region and socioeconomic variation. To overcome such gaps in prognostication, this project is aimed at developing and assessing novel frameworks that can be employed in early post-injury care. Specifically, we

leverage non-parametric models and machine learning techniques to fuse and incorporate routine multimodal and multiplex magnetic resonance imaging (MRI) signals into a prediction framework. In two aims, we address the ability of univariate multimodal fusion and machine learning architectures, respectively, to predict accurately

functional outcome at six months post injury on the sole basis of acute data, and compare their performance to existing clinical algorithms. This project is thus aimed at developing a novel tools that can be easily deployed in the Intensive Care Unit to help guide medical decision-making in an evidence-based manner. If the development

and assessment proposed in the present project is successful, the ultimate aim of this line of work is to develop this research into broadly accessible platform that can be used by practicing clinicians all over the world to supplement prognosis based solely on gross clinical indicators with quantitative and spatial multimodal MR data.