Machine learning-based MRI radiomics identifies patients with degenerative cervical myelopathy and predicts baseline...

Area of Science:

• Neuroimaging
• Radiology
• Machine Learning

Background:

• Degenerative Cervical Myelopathy (DCM) diagnosis relies on clinical evaluation and MRI, but early symptoms are subtle, leading to diagnostic uncertainty.
• Asymptomatic spinal cord compression in older adults complicates surgical intervention decisions.
• MRI-based radiomics offers quantitative texture analysis for microstructural spinal cord pathology, serving as a potential objective imaging biomarker.

Purpose of the Study:

• To assess machine learning models' accuracy in classifying DCM using MRI radiomics.
• To evaluate the models' ability to predict DCM disease severity.
• To determine the impact of imaging resolution on diagnostic performance.

Main Methods:

• Included 79 DCM patients and 51 healthy controls who underwent high-resolution 3D T2-weighted MRI.
• Automated spinal cord segmentation and radiomic feature extraction using Pyradiomics, followed by feature filtering.
• Machine learning algorithms (XGBoost, CatBoost, LightGBM, Random Forest, SVM) were optimized and evaluated using 5-fold cross-validation and an internal test set.

Main Results:

• Machine learning models accurately discriminated DCM from healthy controls (AUROC=0.93, accuracy=0.88).
• Radiomic features from high-resolution MRI outperformed standard MRI and conventional morphometry in diagnostic performance.
• Models effectively classified disease severity (macro-average AUROC=0.855), surpassing traditional imaging measures.

Conclusions:

• MRI-based radiomics combined with machine learning accurately predicts DCM status and disease severity.
• This approach shows potential as an objective imaging biomarker for DCM screening, independent of clinical data.
• External validation in larger, multi-vendor datasets is necessary for broader clinical application.