Value of MRI-based radiomics models for predicting visual outcome in patients with MOG-optic neuritis: A preliminary study

Abstract

Purpose

To investigate the predictive value of MRI-based radiomics models for the recovery of visual acuity after 12 months in patients with acute phase MOG-optic neuritis(MOG-ON).

Materials and Methods

Clinical and MRI imaging data were collected consecutively from January 2021 to April 2022 from patients with acute stage MOG-ON, and the visual acuity of patients were followed up after 12 months. After stratified random sampling, patients were divided into training and test sets, and prediction models based on CE-T1WI, FS-T2WI, and combined CE-T1WI and FS-T2WI were developed.

Results

A total of 34 eyes were included, the visual acuity recovery group included 26 eyes, and the visual acuity non-recovery group included 8 eyes. The training set included 23 eyes (18 from the visual acuity recovery group and 5 from the visual acuity non-recovery group) and the test set included 11 eyes (8 from the visual acuity recovery group and 3 from the visual acuity non-recovery group). The differences in gender, age, proportion of patients with visual acuity recovery, and lesion length between the training and test sets were not statistically significant(P＜0.05). The CE-T1WI-based radiomics model had highest predictive efficacy for visual acuity recovery after 12 months in patients with MOG-ON (AUC = 0.917), followed by the combined CE-T1WI and FS-T2WI-based radiomics model (AUC = 0.838), while the FS-T2WI-based radiomics model had the worst predictive efficacy (AUC = 0.583).

Conclusions

MRI-based radiomics models have high value in predicting visual acuity recovery after 12 months in patients with acute stage MOG-ON, especially CE-T1WI.