Article
Artificial intelligence for the differentiation between multiple sclerosis and glioma II° or III° using 18F-FET-PET imaging
Differenzierung von Multipler Sklerose und Gliomen WHO II°/III° anhand von AI-basierten 18F-FET-PET Untersuchungen
Search Medline for
Authors
Published: | June 4, 2021 |
---|
Outline
Text
Objective: To evaluate the diagnostic significance of 18F-FET-PET imaging for the differentiation between multiple sclerosis (MS) and World Health Organization (WHO) grade 2/3 glioma (glioma II°/III°).
Methods: We retrospectively screened for patients in whom 18F-FET-PET imaging was performed for the diagnostic workup of newly-diagnosed lesions evident on magnetic resonance imaging (MRI) and suspicious for glioma. Among those, we identified patients with histologically confirmed glioma II°/III° and those who later turned out to have multiple sclerosis pursuant to the revised McDonald criteria from 2017. For each group, the mean and maximum tumor-to-brain ratio (TBR) of 18F-FET were determined. Moreover, we used a support-vector machine (SVM) based machine learning algorithm trained on a development and evaluated on a validation subgroup. Receiver operating characteristic (ROC) analysis with area under the curve (AUC) metric was used to assess model performance.
Results: A total of n=33 patients met inclusion criteria. Subsequent diagnostics confirmed MS in n=7 and glioma II°/III° in n=26 patients. TBRmean and TBRmax were significantly higher in the glioma group (TBRmean glioma group: 2.16±0.93, MS group: 1.33±0.14, p = 0.03; TBRmax glioma group: 2.01±0.79, MS group: 1.23±0.19, p = 0.02). In a subgroup analysis, TBRmean and TBRmax significantly differentiated between MS and oligodendroglioma (OG) II°/III° (TBRmean OG group: 2.75±0.91, MS group: 1.33±0.14, p = 0.002; TBRmax OG group: 2.47±0.71, TBRmean MS group: 1.23±0.19, p = 0.003). As shown on ROC analysis, the ability to differentiate between MS and glioma increased from 78.6% using standard TBR analysis to 93.0% using a SVM based machine learning algorithm.
Conclusion: 18F-FET-PET imaging using an SVM based machine learning algorithm enhanced classification performance for the differentiation of MS from glioma. Future studies with larger sample sizes are needed to confirm this observation.