Abstract Neurofibromatosis type 1-associated optic pathway gliomas (NF1-OPG) develop along the anterior visual pathway (AVP: optic nerves, chiasm, and tracts). Prior work suggests that children undergoing treatment for NF1-OPG have larger AVP volumes than those managed with observation, however local MRI radiomic features have not been explored. We developed a fully automated deep-learning framework to segment AVP structures and extract local MRI radiomic features. In this study, we investigate how these radiomic features differ between children who were observed versus treated at enrollment (i.e., time of OPG diagnosis) in the NF1-OPG Natural History Study. Children with newly diagnosed NF1-OPG were included if they had T1-weighted, T2-weighted and T2-FLAIR MRI sequences at study enrollment. AVP volumetric segmentation was performed automatically using a knowledge-transferred Swin Transformer model. Optic nerves and chiasm were subdivided through template-based registration to enable local feature extraction. After histogram matching and Z-score normalization of MRI scans, 1,172 radiomic features were computed. Group differences (observation vs. treatment at time of OPG diagnosis) were assessed using linear mixed-effects modeling. Both MRI shape and appearance features differed significantly (p 0.05) between the observation (OBS, N = 44) and treatment (TX, N = 9) groups. Significant appearance effects included increased T1 texture complexity within the AVP and optic nerves only, and T2 features reflecting small hyperintensity regions throughout the AVP and intensity-distribution asymmetry within the chiasm. At baseline, OBS exhibited significantly higher values in T1 texture complexity (AVP: 5.015 vs. 4.717; right optic nerve: 4.995 vs. 4.736). TX showed significantly greater increases in volumes (AVP: 2.025 vs. 4.605 ml; left optic nerve: 0.563 vs. 1.004; right optic nerve: 0.535 vs. 1.028; chiasm: 0.638 vs. 2.219). These findings suggest that these radiomic features may reflect underlying biological differences or treatment decisions. Automated, local MRI radiomics identified imaging features that may be relevant for treatment decisions in NF1-OPG. This framework, along with future longitudinal analysis, may help support risk stratification for visual acuity loss, predict treatment response and inform individualized monitoring and treatment planning.
Jiang et al. (Tue,) studied this question.