Zhang, Gemeng

Purpose: This study examined whether phoneme-level posterior probabilities generated by an articulatory-informed deep learning model (Phonet) can serve as interpretable and sensitive markers of articulatory impairment in motor speech disorders (MSDs). The goal was to determine whether these features distinguish individuals with and without MSDs and reflect varying severity levels of articulatory impairment. Method: A total of 676 participants (199 with MSDs, 477 without MSDs) completed a remotely captured word repetition task. Speech recordings were phonetically aligned, and posterior probabilities were extracted using a retrained version of the Phonet model, which estimates frame-level probabilities for 23 phonological classes. Segment-level posterior values were computed and compared across groups and severity levels using linear mixed-effects regression models. Results: Posterior probabilities were significantly reduced in the MSD group across all phoneme types. Within the MSD group, posterior probabilities decreased progressively with greater clinical severity, as measured by both perceptual sound distortion ratings and a cumulative articulation error score. Segment-specific analyses revealed systematic differences across phoneme classes, indicating that posterior probabilities capture distinct patterns of articulatory degradation associated with motor speech impairment. Conclusions: Phoneme-level posterior probabilities serve as interpretable, segment-specific indicators of articulatory impairment in MSDs. These features differentiate clinical status and reflect gradations of severity, making them promising tools for augmenting traditional perceptual speech assessments. Their linguistic transparency and scalability support future clinical applications in diagnosis, monitoring, and targeted intervention.

Abstract Functional magnetic resonance imaging (fMRI) data are inherently complex, characterized by high dimensionality, intricate inter-regional dependencies, and substantial individual variability across experimental paradigms. Traditional linear mixed models (LMMs) provide a principled framework that models population-level fixed effects while estimating variance components arising from subject-level random effects; however, they often fail to adequately capture nonlinear relationships inherent in neuroimaging data. To address these limitations, we introduce the nonlinear mixed model (NMM) approach, an innovative extension of the LMM framework that integrates neural networks to flexibly model complex fixed-effect relationships while preserving the random-effects structure to account for individual differences. NMM advances fMRI analysis by: (1) identifying robust functional connectivity (FC) patterns consistently observed across multiple paradigms; (2) leveraging SHapley Additive exPlanations (SHAP) analysis to provide post-hoc interpretability of the nonlinear fixed effects, quantifying how age, sex, and paradigm contribute to predicted FC and how these effects are distributed across large-scale brain networks; and (3) using subject-specific random effects as neural fingerprints that not only show systematic variability across attention and default mode systems but also predict standardized cognitive scores, demonstrating biological relevance. Applied to the Philadelphia Neurodevelopmental Cohort (PNC) across emotion, n-back, and resting-state paradigms, NMM achieved superior model fit relative to classical LMMs, as evidenced by lower mean squared error (MSE) in predicting FC. This framework offers a statistically rigorous and practically explainable approach for modeling large-scale FC from modest covariates while explicitly separating population-level effects from stable individual variability in functional brain organization.

Abstract Background Neuropathologic evaluation is essential for diagnosing and treating dementia, but definitive diagnoses is only available after post‐mortem brain autopsy. Positron emission tomography (PET) offers a potential in vivo proxy for predicting neuropathology. This study proposes a semi‐supervised learning (SSL) model using a single FDG‐PET image to predict underlying neuropathology. Method The SSL model leverages a self‐organized graph structure based on a patient similarity network (PSN) created from 1,495 FDG‐PET images across the aging and neurodegenerative spectrum. Each patient is represented by a node, with edges indicating image similarity derived from principal component analysis features and cosine similarity weights. The PSN structure connects 20 nearest neighbors. Neuropathologic labels were available for 204 autopsied patients. The Poisson Learning Method was employed to propagate labels from these annotated nodes to unlabeled nodes, guided by the PSN structure. FDG‐PET images were collected from clinical and research participants at the Mayo Clinic and processed using an MRI‐free pipeline. Neuropathologic labels were assigned based on widely accepted diagnostic criteria. Leave‐one‐out cross‐validation (LOOCV) was used to assess model performance. Results The confusion matrix demonstrated high accuracy in predicting Alzheimer's disease, frontotemporal lobar degeneration with tau inclusions, and Lewy body disease pathology. A Sankey diagram and label distribution analysis revealed strong correlations between clinical and predicted neuropathologic labels. Conclusion This study highlights the potential of a graph‐based SSL pipeline to predict neuropathological changes using FDG‐PET imaging data, despite limited pathology labels. The findings underscore the value of SSL in bridging the gap between abundant neuroimaging data and scarce neuropathologic labels, contributing to improved dementia diagnosis and understanding.