Multi-parametric MRI phenotype with trustworthy machine learning for differentiating CNS demyelinating diseases

Area of Science:

• Neuroimaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Misdiagnosis of multiple sclerosis (MS) and neuromyelitis optica (NMO) can lead to delayed treatment and poor outcomes.
• Distinguishing between MS and NMO presents clinical challenges.
• Quantitative radiomic features from brain white matter lesions may aid in differential diagnosis.

Purpose of the Study:

• To evaluate the diagnostic value of quantitative radiomic features from brain white matter lesions for differentiating MS and NMO.
• To develop and validate a model for distinguishing between MS and NMO using radiomic and clinical data.

Main Methods:

• Recruited 116 patients with CNS demyelinating diseases (78 MS, 38 NMO).
• Extracted radiomic features from brain white matter lesions on T1-MPRAGE and T2 MRI sequences.
• Developed a Multi-parametric Multivariate Random Forest (MM-RF) model incorporating radiomic and clinical features, validated with cross-validation and independent testing.

Main Results:

• The MM-RF model achieved high accuracy (0.849) and AUC (0.826) in 10-fold cross-validation on the training set.
• In independent testing, the MM-RF model demonstrated excellent performance with AUC 0.902, accuracy 0.871, sensitivity 0.873, and specificity 0.869.
• Clinical factors like age, sex, and EDSS showed mild correlation with radiomic features.

Conclusions:

• Multi-parametric radiomic features show potential as quantitative imaging biomarkers for differentiating MS from NMO.
• This approach can improve diagnostic accuracy and potentially lead to earlier and more appropriate treatment decisions.