Severity Prediction of Traumatic Cervical Spinal Cord Injury With an AI Model Based on MRI Radiomics

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spinal Cord Injury Research

Background:

• Traumatic cervical spinal cord injury (TCSCI) often results in paralysis.
• Current diagnosis relies on subjective MRI interpretation and the American Spinal Injury Association Impairment Scale (AIS) grade.
• There is a need for quantitative markers to accurately assess TCSCI severity.

Purpose of the Study:

• To develop an artificial intelligence (AI) pipeline for predicting AIS grades in TCSCI patients.
• To utilize radiomic features extracted from MRI scans for injury severity assessment.
• To provide a quantitative tool to aid in clinical diagnosis and decision-making.

Main Methods:

• An AI model was developed using MRI data from 130 TCSCI patients for image segmentation.
• Radiomic features were extracted from manually delineated volumes of interest (VOIs) on T2-weighted sagittal images.
• An ensemble model (Em-En) was constructed using selected radiomic features for AIS grade prediction.

Main Results:

• The UCTransnet and U-Net++ based segmentation model achieved high performance (mDICE: 0.777, mIOU: 0.646).
• The Em-En model demonstrated superior performance in predicting AIS grades compared to other models.
• The AI pipeline showed potential for improving sensitivity, specificity, and accuracy in clinical decision-making.

Conclusions:

• An AI-assisted pipeline was successfully developed for predicting TCSCI severity.
• This AI pipeline offers a theoretical foundation for clinical application, potentially reducing MRI interpretation barriers.
• The developed resources provide quantitative indicators for injury severity, aiding clinicians in diagnosis and treatment planning.