MACHINE LEARNING FOR IDENTIFYING INJURED ELEMENTS IN COMPUTATIONAL MODELS OF SPINAL CORD INJURY

Abstract

This study leverages machine learning (ML) algorithms to identify tissue damage in computational models of 
spinal cord injury (SCI) based on mechanical outputs. Three datasets—corresponding to gray matter, white 
matter, and their combination—were constructed from comparisons between histological images from SCI 
experiments in non-human primates and subject-specific finite element (FE) models. Four ML algorithms were 
assessed using cross-validation and the area under the receiver operating characteristic curve (AUC) metric. 
Following hyperparameter optimization, AUC mean values ranged from 0.79 to 0.82, with a standard deviation 
no greater than 0.02. Among the algorithms, k-nearest neighbors and logistic regression demonstrated superior 
performance in identifying injured elements compared to support vector machines and decision trees.  
The results contribute to understanding the relationship between mechanical loading and tissue damage in SCI, 
with implications for developing prevention strategies. Clinical Relevance: By linking FE model predictions to 
tissue damage, this approach enhances the clinical utility of FE models. Combined with imaging technologies, 
these models can predict damage extent in animal studies and inform treatment planning decisions