Linking speech patterns to brain structure in affective and psychotic disorders: an integrative natural language processing approach

作者：Svenja Seuffert, Rieke Roxanne Mülfarth, Lea Teutenberg, Florian Thomas‐Odenthal, Paula Usemann, Nina Alexander, Hamidreza Jamalabadi, Igor Nenadić, Benjamin Straube, Tim Hahn, Udo Dannlowski, Tilo Kircher, Frederike Stein · 发表于：Molecular Psychiatry · 年份：2025 · DOI：10.1038/s41380-025-03347-9 · 被引用次数：3 · 研究领域：Schizophrenia research and treatment、Advanced Neuroimaging Techniques and Applications、Stuttering Research and Treatment

Language disturbances are central features of serious mental illnesses, yet traditional clinical assessments often rely on subjective evaluation that may overlook subtle speech anomalies. This study employs natural language processing (NLP) to objectively analyze spontaneous speech in a transdiagnostic sample comprising individuals with affective (n = 119 Major Depressive Disorder, n = 27 Bipolar Disorder) and psychotic disorders (n = 37 Schizoaffective Disorder, n = 11 Schizophrenia), as well as healthy controls (n = 178). Participants provided approximately 12 min of speech elicited via four pictures from the Thematic Apperception Test, which were transcribed and analyzed for semantic and syntactic features. Explorative factor analysis identified three latent linguistic dimensions: (1) Syntactic Complexity, (2) Lexical Diversity and Fluency, and (3) Narrow Thematic Focus. These dimensions were differentially associated with clinical ratings of formal thought disorder and neuroanatomical measures obtained through structural and diffusion-weighted MRI. Notably, Syntactic Complexity and Lexical Diversity correlated with decreased fractional anisotropy (FA) in frontotemporal white matter tracts, while Narrow Thematic Focus was linked to reduced gray matter volume in the right posterior insula. Importantly, these associations persisted after controlling for diagnosis, medication status, and verbal IQ. These findings suggest that NLP-derived speech metrics can serve as sensitive ...