Abstract
Purpose :
Mild traumatic brain injury (mTBI) is a complex condition and the pathophysiology of mTBI-related visual impairment remains unclear. Variability in visual dysfunction can also delay diagnosis and treatment. The purpose of this study was to investigate the effects of mTBI on the optic radiations and occipital lobe using multimodal analyses and to apply machine learning models to evaluate novel imaging biomarkers.
Methods :
This matched pair case-control study (n=36, age=33±7 years) included mTBI with Glasgow coma scale (GCS) score 13-15 and control groups. Exclusion criteria were non-mTBI vision pathology, TBI with GCS <13, age <18 years old, pregnancy, and metallic implants. T1 and diffusion-weighted (DWI) MRI were acquired during a single session with the same acquisition protocols across participants. 3D whole brain segmentation into 135 regions with the spatially localized atlas network tiles (SLANT) deep learning pipeline was applied on each T1 MRI. DWI scans were preprocessed with PreQual for quality assurance. The optic radiations were extracted from the tractography pipeline (Tractoflow, RecoBundlesX, TractometryFlow). Random forest machine learning models with leave-one-out cross-validation evaluated optic radiations and occipital lobe metrics as potential biomarkers for mTBI-related visual impairment.
Results :
Mean (p=0.010) and minimum (p=0.009) fiber length and total bundle volume (p=0.046) of the optic radiations in mTBI were significantly higher compared to control in paired t-tests. The machine learning model included optic radiations tractography macrostructure (mean fiber length, minimum fiber length, total bundle volume, total streamline counts) and microstructure (mean, axial, and radial diffusivity and fractional anisotropy) metrics and 14 occipital lobe segments, normalized by total brain volume, that are primarily involved in vision. This model (accuracy=0.72, 95% CI [0.58, 0.87], ROC-AUC=0.70) outperformed the null model with age & sex (accuracy=0.41, 95% CI [0.25, 0.57], ROC-AUC=0.45).
Conclusions :
Multimodal tractography analysis showed differences in the macrostructure of the optic radiations and occipital lobe regions between mTBI and control groups. These findings suggest their potential as novel imaging biomarkers for mTBI-related visual impairment.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.