Quantitative brain imaging analysis of neurological syndromes associated with anti-GAD antibodies

Neurological disorders associated with anti-glutamic acid decarboxylase (GAD) autoimmunity are rare and include a variety of neurological syndromes: stiff-person syndrome, cerebellar ataxia or limbic encephalitis. The diagnosis remains challenging due to the variety of symptoms and normal brain imaging. The morphological MRI of 26 patients (T1-weighted and Fluid-attenuated inversion recovery (FLAIR) weighted images) was analyzed at the initial stage of diagnosis, matched by age and sex to 26 healthy subjects. We performed a vertex-wise analysis using a generalized linear model, adjusting by age, to compare the brain cortical thickness of both populations. In addition, we used a voxel-based morphometry of cerebellum thickness obtained by CEREbellum Segmentation (CERES), as well as the hippocampus volumetry comparison using HIPpocampus subfield Segmentation (HIPS). Finally, we extracted 62 radiomics features using LifeX to assess the classification performance using a random forest model to identify an anti-GAD related MRI. The results suggest a peculiar profile of atrophy in patients with anti-GAD, with a significant atrophy in the temporal and frontal lobes (adjusted p-value < 0.05), and a focal cerebellar atrophy of the V-lobule, independently of the anti-GAD phenotype. Finally, the MRIs from anti-GAD patients were correctly classified when compared to the control group, with an area under the curve (AUC) of 0.98. This study suggests a particular pattern of cortical atrophy throughout all anti-GAD phenotypes. These results reinforce the notion that the different neurological anti-GAD phenotypes should be considered as a continuum due to their similar cortical thickness profiles.

Automated summary

Patients with anti-GAD show a unique pattern of cortical atrophy in MRI, particularly in the temporal and frontal lobes as well as the V-lobule of the cerebellum. The MRIs of anti-GAD patients were accurately classified compared to controls, with a high AUC value, using a random forest model.