Machine learning-based differentiation between multiple sclerosis and glioma WHO II°-IV° using O-(2-[18F] fluoroethyl)-L-tyrosine positron emission tomography.

INTRODUCTION

This study aimed to test the diagnostic significance of FET-PET imaging combined with machine learning for the differentiation between multiple sclerosis (MS) and glioma II°-IV°.

METHODS

Our database was screened for patients in whom FET-PET imaging was performed for the diagnostic workup of newly diagnosed lesions evident on MRI and suggestive of glioma. Among those, we identified patients with histologically confirmed glioma II°-IV°, and those who later turned out to have MS. For each group, tumor-to-brain ratio (TBR) derived features of FET were determined. A support vector machine (SVM) based machine learning algorithm was constructed to enhance classification ability, and Receiver Operating Characteristic (ROC) analysis with area under the curve (AUC) metric served to ascertain model performance.

RESULTS

A total of 41 patients met selection criteria, including seven patients with MS and 34 patients with glioma. TBR values were significantly higher in the glioma group (TBRmax glioma vs. MS: p = 0.002; TBRmean glioma vs. MS: p = 0.014). In a subgroup analysis, TBR values significantly differentiated between MS and glioblastoma (TBRmax glioblastoma vs. MS: p = 0.0003, TBRmean glioblastoma vs. MS: p = 0.0003) and between MS and oligodendroglioma (ODG) (TBRmax ODG vs. MS: p = 0.003; TBRmean ODG vs. MS: p = 0.01). The ability to differentiate between MS and glioma II°-IV° increased from 0.79 using standard TBR analysis to 0.94 using a SVM based machine learning algorithm.

CONCLUSIONS

FET-PET imaging may help differentiate MS from glioma II°-IV° and SVM based machine learning approaches can enhance classification performance.