Determination of Diagnosis and Prognosis in Spinal Cord Injury Using Machine Learning

Abstract

Traumatic spinal cord injury (tSCI) is a neurological disorder that leads to long-term disability, significant economic burden, and limited treatment options. Accurate and timely diagnosis, as well as reliable prognosis, are crucial for acute treatment, rehabilitation, and long-term treatment planning. Recent advances in artificial intelligence (AI) and machine learning have demonstrated significant potential to support the diagnostic workflow and predict clinical outcomes in patients with tSCI. For diagnostic purposes, models based on convolutional neural networks trained using magnetic resonance imaging and diffusion tensor imaging showed high accuracy in detecting cord damage and classifying injury severity. A variety of prognostic models, ranging from traditional logistic regression to advanced neural networks and deep learning-based radiomics, have been applied to predict functional recovery, ambulatory status, and survival. Although AI-based approaches generally have better prediction accuracy than conventional methods, several limitations remain, such as limited dataset sizes, heterogeneity across studies, lack of external validation. Future studies should incorporate multicenter collaborations, standardized reporting frameworks, and integration of multimodal data to enhance clinical generalizability and applicability. With further improvements, AI will play a crucial role in supporting clinicians in making decisions about tSCI and in patient rehabilitation.