Late concussion recovery prediction

The goal of this project was to analyze clinical and neuroimaging data of concussed athletes for late concussion recovery prediction.

A concussion is a form of mild Traumatic Brain Injury (mTBI) that leads to temporary alterations of brain functions and incapacitates behavior. Although in most of the cases the symptoms resolve within a few days to a couple of weeks, sometimes they can last for months or even for years, a condition identified as late recovery. Early identification of late recoveries would allow for early and better optimized treatment.

The CARE dataset, distributed by the NCAA/DoD Concussion Assessment Research and Education (CARE) consortium, is the largest open dataset currently available for concussion research with over 35,000 among student and military cadet athletes. Only concussed participants with data collected within 24-48h from injury were analyzed. They were divided into two standard groups based on their recovery time: (1) the early recovery group (recovery time < 28 days), and (2) the late recovery group (otherwise).

CLINICAL DATA ANALYSIS

Machine learning methods were used to classify the participants into the two recovery groups. Multiple classifiers and combinations of variables as input were tested using Repeated Stratified 5-Fold Cross Validation. Their performances were evaluated using multiple classification metrics on a held-out test set.

NEUROIMAGING DATA ANALYSIS

A cutting edge Magnetic Resonance Imaging (MRI) protocol was applied to a subset of the participants by the CARE consortium, comprising anatomical and diffusion weighted MRI data. A fully reproducible neuroimaging pipeline was implemented on brainlife. Specifically, tractography, white matter tract segmentation (47 tracts), and tract Fractional Anisotropy (FA) profile analysis were performed. After computing the mean FA tract profile for each of the 47 tracts, a Logistic Regression (LR) binary classifier was used to classify late concussion recoveries. Its performance was then evaluated using multiple classification metrics on a held-out test set. An ROC AUC score of 0.90 was obtained, with sensitivity=1, and specificity=0.79.

Read the full paper here: https://doi.org/10.1016/j.nicl.2024.103646
Read the poster here: ppcs-poster.pdf
Try the brainlife App here: https://doi.org/10.25663/brainlife.app.793
Code: app-predict-ppcs

Citation: Bertò, G., Rooks, L. T., Broglio, S. P., McAllister, T. A., McCrea, M. A., Pasquina, P. F., ... & Pestilli, F. (2024). Diffusion tensor analysis of white matter tracts is prognostic of persisting post-concussion symptoms in collegiate athletes. NeuroImage: Clinical, 103646.