A 2026 medRxiv preprint found that REDDI, a resting-state magnetoencephalography machine-learning pipeline, separated mild cognitive impairment, multiple sclerosis, Parkinson’s disease, and amyotrophic lateral sclerosis with 0.81 ± 0.04 balanced accuracy across 5 folds.1 That is a meaningful jump over the prior 67.1% MEG benchmark, but it is still research-stage decision support rather than a clinical …
A 2026 machine-learning analysis of 110 adults with generalized anxiety disorder predicted 2-week digital mindfulness engagement with R2 = 82.1% in a top-10 predictor model, and the model favored mindfulness prompts over self-monitoring prompts for engagement (d = 1.447, p < .001).1 The calibrated read is that engagement with brief app-based mindfulness may be matchable, …
A 2026 ADNI study reported 85.23% accuracy for ANA-GNN, a graph neural network that combined structural MRI regional features with clinical variables to classify cognitively normal controls, mild cognitive impairment, and Alzheimer’s disease.1 The result is useful, but the clinical-feature ablation dropped accuracy to 68.35%, so the model should be read as multimodal decision-support research, …
A 2026 serum multiomics study linked occupational noise exposure to tinnitus severity mostly through metabolism: 10 metabolites, including GABA, fumaric acid, and steroid hormone precursors, statistically mediated 92% of the exposure-tinnitus association.1 The result is not a clinical blood test yet, but it pushes tinnitus biology beyond the ear-only frame toward a metabolism-immunity model that …
A 2026 Scientific Reports study of Maryland suicide-death records found that an XGBoost machine-learning model could reach 96.1% positive predictive value in hospital-discharge data at the top 0.1% risk threshold, but it still detected only 46.7% of suicide deaths in that cohort. Research Highlights Precision improved at the narrowest threshold: XGBoost reached PPV 0.961 in …
A 2026 multimodal MRI study found that treatment-resistant depression had lower coordination between brain structure and brain activity than non-treatment-resistant depression in frontal, parietal, motor, and temporal regions. Machine-learning models using those structure-function coupling measures classified treatment-resistant vs. non-treatment-resistant depression with AUC values from 0.886 to 0.950 in the primary atlas analysis.1 Research Highlights TRD …
A 2026 diagnostic-accuracy preprint found that resting-state EEG network features separated non-lesional epilepsy from functional/dissociative seizures (FDS) at 67.5% balanced accuracy, but the signal was not symmetrical: sensitivity was 81.8% for epilepsy and only 53.3% for FDS.1 In plain clinical terms, the model looked more useful as an epilepsy-supporting marker than as a positive test …