Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.
Microstructural alterations have been reported in patients with urologic chronic pelvic pain syndrome (UCPPS). However, it isn't clear whether these alterations are reproducible within 6 months or whether long-term symptom improvement is associated with specific microstructural changes. Using data from the MAPP-II Research Network, the current study performed population-based voxel-wise DTI and probabilistic tractography in a large sample of participants from the multicenter cohort with UCPPS (N = 364) and healthy controls (HCs, N = 61) over 36 months. While fractional anisotropy (FA) differences between UCPPS patients and HCs were observed to be unique at baseline and 6-month follow-up visits, consistent aberrations in mean diffusivity (MD) were observed between UCPPS and HCs at baseline and repeated at 6 months. Additionally, compared to HCs, UCPPS patients showed stronger structural connectivity (SC) between the left postcentral gyrus and the left precuneus, and weaker SC from the left cuneus to the left lateral occipital cortex and the isthmus of the left cingulate cortex at baseline and 6-month. By 36 months, reduced FA and MD aberrations in these same regions were associated with symptom improvement in UCPPS. Together, results suggest changes in white matter microstructure may play a role in the persistent pain symptoms in UCPPS. PERSPECTIVE: This longitudinal study identified reproducible, "disease-associated" patterns in altered mean diffusivity and abnormal microstructural connectivity in UCPPS comparing to HCs over 6 months. These differences were found in regions involved in sensory processing and integration and pain modulation, making it potentially amenable for clinical interventions that target synaptic and/or neuronal reorganization.
Characterization of hypoxia and tissue acidosis could advance the understanding of glioma biology and improve patient management. In this study, we evaluated the ability of a pH- and oxygen-sensitive magnetic resonance imaging (MRI) technique to differentiate glioma genotypes, including isocitrate dehydrogenase (IDH) mutation, 1p/19q co-deletion, and epidermal growth factor receptor (EGFR) amplification, and investigated its prognostic value. A total of 159 adult glioma patients were scanned with pH- and oxygen-sensitive MRI at 3T. We quantified the pH-sensitive measure of magnetization transfer ratio asymmetry (MTRasym) and oxygen-sensitive measure of R2’ within the tumor region-of-interest. IDH mutant gliomas showed significantly lower MTRasym × R2’ (p < 0.001), which differentiated IDH mutation status with sensitivity and specificity of 90.0% and 71.9%. Within IDH mutants, 1p/19q codeletion was associated with lower tumor acidity (p < 0.0001, sensitivity 76.9%, specificity 91.3%), while IDH wild-type, EGFR-amplified gliomas were more hypoxic (R2’ p = 0.024, sensitivity 66.7%, specificity 76.9%). Both R2’ and MTRasym × R2’ were significantly associated with patient overall survival (R2’: p = 0.045; MTRasym × R2’: p = 0.002) and progression-free survival (R2’: p = 0.010; MTRasym × R2’: p < 0.0001), independent of patient age, treatment status, and IDH status. The pH- and oxygen-sensitive MRI is a clinically feasible and potentially valuable imaging technique for distinguishing glioma subtypes and providing additional prognostic value to clinical practice.
Degenerative cervical myelopathy (DCM) is a progressive condition characterized by degeneration of osseocartilaginous structures within the cervical spine resulting in compression of the spinal cord and presentation of clinical symptoms. Compared to healthy controls (HCs), studies have shown DCM patients experience structural and functional reorganization in the brain; however, sex-dependent cortical differences in DCM patients remains largely unexplored. In the present study, we investigate the role of sex differences on the structure of the cerebral cortex in DCM and determine how structural differences may relate to clinical measures of neurological function. T1-weighted structural MRI scans were acquired in 85 symptomatic and asymptomatic patients with DCM and 90 age-matched HCs. Modified Japanese Orthopedic Association (mJOA) scores were obtained for patients. A general linear model was used to determine vertex-level significant differences in gray matter volume (GMV) between the following groups (1) male HCs and female HCs, (2) male patients and female patients, (3) male patients and male HCs, and (4) female patients and female HCs. Within patients, males exhibited larger GMV in motor, language, and vision related brain regions compared to female DCM patients. Males demonstrated a significant positive correlation between GMV and mJOA score, in which patients with worsening neurological symptoms exhibited decreasing GMV primarily across somatosensory and motor related cortical regions. Females exhibited a similar association, albeit across a broader range of cortical areas including those involved in pain processing. In sensorimotor regions, female patients consistently showed smaller GMV compared with male patients, independent of mJOA score. Results from the current study suggest strong sex-related differences in cortical volume in patients with DCM, which may reflect hormonal influence or differing compensation mechanisms.
This longitudinal study evaluated the cortical thickness, gray-to-white matter contrast (GWC), and frontal lobe intracortical myelin (ICM) volume in first-episode schizophrenia (FES) patients treated with oral antipsychotics (OAP) versus a long-acting injectable antipsychotic, paliperidone palmitate (PP). 2D proton density and inversion recovery images, and 3D T1-weighted MPRAGE images were acquired at 3T from 68 FES patients in a randomized clinical trial with PP vs OAP. At baseline, no differences in GWC and ICM were observed between FES patients and HCs, but the thickness of the left precuneus, the right transverse temporal gyrus, and the bilateral superior temporal gyri was found to be thinner in FES patients relative to HCs. Following 9 months of antipsychotics, OAP treatment, compared to PP treatment, resulted in a more widespread cortical thickness reduction including the right lateral occipital and orbitofrontal gyri. No significant ICM and GWC changes were observed in the PP group, whereas OAP treatment led to a significant ICM volume decrease and GWC increase. A negative correlation was found between ICM changes and GWC changes within multiple frontal regions after 9 months of OAP treatment. These preliminary findings suggest that PP treatment might aid preservation of brain morphology.