Machine learning can be used to define subtypes of psychiatric conditions based on shared biological foundations of mental disorders. Here we analyzed cross-sectional brain images from 4,222 individuals with schizophrenia and 7038 healthy subjects pooled across 41 international cohorts from the ENIGMA, non-ENIGMA cohorts and public datasets. Using the Subtype and Stage Inference (SuStaIn) algorithm, we identify two distinct neurostructural subgroups by mapping the spatial and temporal ‘trajectory’ of gray matter change in schizophrenia. Subgroup 1 was characterized by an early cortical-predominant loss with enlarged striatum, whereas subgroup 2 displayed an early subcortical-predominant loss in the hippocampus, striatum and other subcortical regions. We confirmed the reproducibility of the two neurostructural subtypes across various sample sites, including Europe, North America and East Asia. This imaging-based taxonomy holds the potential to identify individuals with shared neurobiological attributes, thereby suggesting the viability of redefining existing disorder constructs based on biological factors.

BACKGROUND AND PURPOSE: Deep gray matter involvement is a consistent feature in multiple sclerosis. The aim of this study was to evaluate the relationship between different deep gray matter alterations and the development of subcortical atrophy, as well as to investigate the possible different substrates of volume loss between phenotypes. MATERIALS AND METHODS: Seventy-seven patients with MS (52 with relapsing-remitting and 25 with progressive MS) and 41 healthy controls were enrolled in this cross-sectional study. MR imaging investigation included volumetric, DTI, PWI and Quantitative Susceptibility Mapping analyses. Deep gray matter structures were automatically segmented to obtain volumes and mean values for each MR imaging metric in the thalamus, caudate, putamen, and globus pallidus. Between-group differences were probed by ANCOVA analyses, while the contribution of different MR imaging metrics to deep gray matter atrophy was investigated via hierarchic multiple linear regression models. RESULTS: Patients with MS showed a multifaceted involvement of the thalamus and basal ganglia, with significant atrophy of all deep gray matter structures (P < .001). In the relapsing-remitting MS group, WM lesion burden proved to be the main contributor to volume loss for all deep gray matter structures (P ≤ .006), with a minor role of local microstructural damage, which, in turn, was the main determinant of deep gray matter atrophy in patients with progressive MS (P ≤ .01), coupled with thalamic susceptibility changes (P = .05). CONCLUSIONS: Our study confirms the diffuse involvement of deep gray matter in MS, demonstrating a different behavior between MS phenotypes, with subcortical GM atrophy mainly determined by global WM lesion burden in patients with relapsing-remitting MS, while local microstructural damage and susceptibility changes mainly accounted for the development of deep gray matter volume loss in patients with progressive MS.

BACKGROUND AND PURPOSE: Modifications of magnetic susceptibility have been consistently demonstrated in the subcortical gray matter of MS patients, but some uncertainties remain concerning the underlying neurobiological processes and their clinical relevance. We applied quantitative susceptibility mapping and longitudinal relaxation rate relaxometry to clarify the relative contribution of atrophy and iron and myelin changes to deep gray matter damage and disability in MS. MATERIALS AND METHODS: Quantitative susceptibility mapping and longitudinal relaxation rate maps were computed for 91 patients and 55 healthy controls from MR images acquired at 3T. Applying an external model, we estimated iron and myelin concentration maps for all subjects. Subsequently, changes of deep gray matter iron and myelin concentration (atrophy-dependent) and content (atrophy-independent) were investigated globally (bulk analysis) and regionally (voxel-based and atlas-based thalamic subnuclei analyses). The clinical impact of the observed MRI modifications was evaluated via regression models. RESULTS: We identified reduced thalamic (P < .001) and increased pallidal (P < .001) mean iron concentrations in patients with MS versus controls. Global myelin and iron content in the basal ganglia did not differ between the two groups, while actual iron depletion was present in the thalamus (P < .001). Regionally, patients showed increased iron concentration in the basal ganglia (P ≤ .001) and reduced iron and myelin content in thalamic posterior-medial regions (P ≤ .004), particularly in the pulvinar (P ≤ .001). Disability was predicted by thalamic volume (B = –0.341, P = .02), iron concentration (B = -0.379, P = .005) and content (B = –0.406, P = .009), as well as pulvinar iron (B = –0.415, P = .003) and myelin (B = −0.415, P = .02) content, independent of atrophy. CONCLUSIONS: Quantitative MRI suggests an atrophy-related iron increase within the basal ganglia of patients with MS, along with an atrophy-independent reduction of thalamic iron and myelin correlating with disability. Absolute depletions of thalamic iron and myelin may represent sensitive markers of subcortical GM damage, which add to the clinical impact of thalamic atrophy in MS.

Objectives: To evaluate the accuracy of a data-driven approach, such as machine learning classification, in predicting disability progression in MS.Methods: We analyzed structural brain images of 163 subjects diagnosed with MS acquired at two different sites. Participants were followed up for 2–6 years, with disability progression defined according to the expanded disability status scale (EDSS) increment at follow-up. T2-weighted lesion load (T2LL), thalamic and cerebellar gray matter (GM) volumes, fractional anisotropy of the normal appearing white matter were calculated at baseline and included in supervised machine learning classifiers. Age, sex, phenotype, EDSS at baseline, therapy and time to follow-up period were also included. Classes were labeled as stable or progressed disability. Participants were randomly chosen from both sites to build a sample including 50% patients showing disability progression and 50% patients being stable. One-thousand machine learning classifiers were applied to the resulting sample, and after testing for overfitting, classifier confusion matrix, relative metrics and feature importance were evaluated.Results: At follow-up, 36% of participants showed disability progression. The classifier with the highest resulting metrics had accuracy of 0.79, area under the true positive versus false positive rates curve of 0.81, sensitivity of 0.90 and specificity of 0.71. T2LL, thalamic volume, disability at baseline and administered therapy were identified as important features in predicting disability progression. Classifiers built on radiological features had higher accuracy than those built on clinical features.Conclusions: Disability progression in MS may be predicted via machine learning classifiers, mostly evaluating neuroradiological features.

To test the performance of the 2023 myelin oligodendrocyte glycoprotein antibody–associated disease (MOGAD) criteria in adults and children with inflammatory demyelinating conditions who were tested for MOG antibodies (Abs). This was a retrospective study of patients tested for MOG-Abs from 2018 to 2022 in 2 specialist hospitals. The inclusion criteria comprised ≥1 attendance in an adult or pediatric demyelinating disease clinic and complete clinical and MRI records. The final clinical diagnosis of MOGAD, made by the treating neurologist, was taken as the benchmark against which the new criteria were tested. The international MOGAD diagnostic criteria were applied retrospectively; they stipulate at least 1 clinical or MRI supporting feature for MOGAD diagnosis in positive fixed MOG cell-based assay without a titer. The performance MOG-Ab testing alone for MOGAD diagnosis was also assessed and compared with that of MOGAD criteria using the McNemar test. Of the 1,879 patients tested for MOG-Abs, 539 (135 pediatric and 404 adults) met the inclusion criteria. A clinical diagnosis of MOGAD was made in 86/539 (16%) patients (37 adults, 49 children), with a median follow-up of 3.6 years. The MOGAD diagnostic criteria had sensitivity of 96.5% (adults 91.9%, children 100%), specificity of 98.9% (adults 98.8%, children 98.9%), positive predictive value of 94.3% (adults 89.4%, children 98%), negative predictive value of 99.3% (adults 99.2%, children 100%), and accuracy of 98.5% (adults 98.3%, children 99.2%). When compared with MOG-Ab testing alone, a difference was seen only in adults: a significantly higher specificity (98.9% vs 95.6%, p = 0.0005) and nonstatistically significant lower sensitivity (91.9% vs 100%, p = 0.08). The international MOGAD diagnostic criteria exhibit high performance in selected patients with inflammatory demyelinating diseases (who had a high pretest probability of having MOGAD) compared with best clinical judgment; their performance was better in children than in adults. In adults, the MOGAD criteria led to an improvement in specificity and positive predictive value when compared with MOG-Ab testing alone, suggesting that the requirement of at least 1 clinical or MRI supporting feature is important. Future work should address the generalizability of the diagnostic criteria to cohorts of greater clinical diversity seen within neurologic settings.