Longitudinal neuroimaging and neurocognitive assessment of risk and protective factors across the schizophrenia spectrum

摘要

ABSTRACT The parent R01 project is a longitudinal study examining risk and protective factors in the schizophrenia (SZ) spectrum—from healthy controls (HCs) to individuals with schizotypal personality disorder (SPD) to recent-onset SZ patients (80 per group)—using MRI and neurocognitive approaches. It tests a neurobiological model which posits that individuals with SPD—an intermediate phenotype—have protective factors against developing threshold psychosis, such as preservation of frontal lobe and less severe temporal lobe abnormalities compared to SZ that lead to milder cognitive and social impairments. Examining natural language processing (NLP) as proposed in this supplement is in line with the scope and aims of the parent R01 project and may inform the key neurobiological model being tested. Moreover, examining NLP using novel measures of semantics and syntax in association with measures from the parent R01 of frontal and temporal white matter integrity/connectivity assessed with diffusion tensor imaging and cognitive domains such as processing speed and working memory is innovative. Speech and language provide a rich source of data on human thought, including semantic and emotional content, semantic coherence (i.e. flow of meaning), and syntactic structure and complexity (i.e. usage of parts of speech). There is a critical gap in our understanding of the linguistic mechanisms that underlie thought disorder in SZ spectrum. The use of automated linguistic analytic methods has been limited to only a few studies focused on discriminating SZ patients from HCs and predicting psychosis. Together with our colleagues with expertise in NLP at Icahn School of Medicine at Mount Sinai, we will use advanced computational speech analytic approaches to identify the linguistic basis of language production along a spectrum from normal to thought disordered. We will use optimal interviewing techniques1 to obtain open-ended 30-45 minute narratives from the large (N = 240) English-speaking sample in the parent R01 study, with a range of language disturbances across the spectrum ranging from none/subtle to severe. NLP techniques including Latent Semantic Analysis2 (LSA) and part-of-speech (POS) tagging3,4 will be conducted using artificial intelligence to examine semantic and syntactic language features to include in our overall neurobiological model. These analyses yield fine-grained indices of speech and language that may more accurately capture thought disorder. Three specific aims will assess (1) semantic coherence in language production using LSA2 and examine its association with positive symptoms and functional impairment across the spectrum; (2) syntactic complexity in language production using POS tagging3,4 and measure acoustic features to examine their association with negative symptoms and functional impairment; and (3) the relationship between language and speech features (semantic, syntactic, and acoustic) with putative white matter integrity assessed using diffusion tensor imaging.