Establishment and validation of a ResNet-based radiomics model for predicting prognosis in cervical spinal cord injury patients

Scientific Reports, 2025 · DOI: https://doi.org/10.1038/s41598-025-94358-7 · Published: March 13, 2025

Simple Explanation

Cervical spinal cord injury (cSCI) has unpredictable recovery, from mild paralysis to severe disability. Accurate prediction models are needed for treatment. This study creates a model using imaging (radiomics, deep learning) and clinical data to predict cSCI prognosis six months after injury. The model combines clinical factors like age, diabetes, and treatment type with radiomic features extracted from MR images using ResNet deep learning.

Study Duration

5 Years, 11 Months

Participants

211 cSCI patients

Evidence Level

Not specified

Key Findings

1
The SVM classifier achieved the highest AUC of 1.000 in the training set and 0.915 in the testing set for radiomics models.
2
Age, diabetes, and treatment were found to be independent clinical risk factors.
3
The combined model, integrating radiomics and clinical features, achieved AUCs of 1.000 in the training set, 0.952 in the testing set, and 0.815 in the validation set.

Research Summary

This study developed a combined radiomics and clinical model to predict the prognosis of cSCI patients six months post-injury. The model integrates radiomic features extracted using Pyradiomics and ResNet deep learning from MR images with clinical factors such as age, diabetes, and treatment type. The combined model demonstrated strong performance with high AUC values in the training, testing, and validation sets, indicating its potential for clinical use.

Practical Implications

Personalized Treatment Plans

The combined model can help in developing personalized treatment plans for cSCI patients based on predicted prognosis.

Improved Clinical Decision-Making

The model can assist clinicians in making more informed decisions regarding patient consultation, treatment, and rehabilitation.

Stratified Prognostic Assessments

The model provides stratified prognostic assessments, which can help guide treatment and rehabilitation decisions.

Study Limitations

1
Traditional MRI struggles to reveal molecular information beneath the macroscopic level, such as axonal and myelin preservation
2
The need for more external independent validation sets to ensure the generalizability of our findings
3
Challenges such as volume effects in spinal cord imaging, metal artifacts, and the impracticality of requiring patients to remain precisely still for long periods at high resolution

Your Feedback

Was this summary helpful?

Back to Spinal Cord Injury