A Comprehensive Review of Computational Methods for Automatic Prediction of Schizophrenia With Insight Into Indigenous Populations

Automated linguistic analysis in speech samples of Turkish-speaking patients with schizophrenia-spectrum disorders

Modern natural language processing (NLP) methods provide ways to objectively quantify language disturbances for potential use in diagnostic classification. We performed computerized language analysis in speech samples of 82 Turkish-speaking subjects, including 44 patients with schizophrenia spectrum disorders (SSD) and 38 healthy controls (HC). Exploratory analysis of speech samples involved 16 sentence-level semantic similarity features using SBERT (Sentence Bidirectional Encoder Representation from Text) as well as 8 generic and 8 part-of-speech (POS) features. The random forest classifier using SBERT-derived semantic similarity features achieved a mean accuracy of 85.6 % for the classification of SSD and HC. When semantic similarity features were combined with generic and POS features, the classifier's mean accuracy reached to 86.8 %. Our analysis reflected increased sentence-level semantic similarity scores in SSD. Generic and POS analyses revealed an increase in the use of verbs, proper nouns and pronouns in SSD while our results showed a decrease in the utilization of conjunctions, determiners, and both average and maximum sentence length in SSD compared to HC. Quantitative language features were correlated with the expressive deficit domain of BNSS (Brief Negative Symptom Scale) as well as with the duration of illness. These findings from Turkish-speaking interviews contribute to the growing evidence-based NLP-derived assessments in non-English-speaking patients.

Active Inference, Epistemic Value, and Uncertainty in Conceptual Disorganization in First-Episode Schizophrenia

BACKGROUND AND HYPOTHESIS

Active inference has become an influential concept in psychopathology. We apply active inference to investigate conceptual disorganization in first-episode schizophrenia. We conceptualize speech production as a decision-making process affected by the latent "conceptual organization"-as a special case of uncertainty about the causes of sensory information. Uncertainty is both minimized via speech production-in which function words index conceptual organization in terms of analytic thinking-and tracked by a domain-general salience network. We hypothesize that analytic thinking depends on conceptual organization. Therefore, conceptual disorganization in schizophrenia would be both indexed by low conceptual organization and reflected in the effective connectivity within the salience network.

STUDY DESIGN

With 1-minute speech samples from a picture description task and resting state fMRI from 30 patients and 30 healthy subjects, we employed dynamic causal and probabilistic graphical models to investigate if the effective connectivity of the salience network underwrites conceptual organization.

STUDY RESULTS

Low analytic thinking scores index low conceptual organization which affects diagnostic status. The influence of the anterior insula on the anterior cingulate cortex and the self-inhibition within the anterior cingulate cortex are elevated given low conceptual organization (ie, conceptual disorganization).

CONCLUSIONS

Conceptual organization, a construct that explains formal thought disorder, can be modeled in an active inference framework and studied in relation to putative neural substrates of disrupted language in schizophrenia. This provides a critical advance to move away from rating-scale scores to deeper constructs in the pursuit of the pathophysiology of formal thought disorder.

Latent Factors of Language Disturbance and Relationships to Quantitative Speech Features

BACKGROUND AND HYPOTHESIS

Quantitative acoustic and textual measures derived from speech ("speech features") may provide valuable biomarkers for psychiatric disorders, particularly schizophrenia spectrum disorders (SSD). We sought to identify cross-diagnostic latent factors for speech disturbance with relevance for SSD and computational modeling.

STUDY DESIGN

Clinical ratings for speech disturbance were generated across 14 items for a cross-diagnostic sample (N = 334), including SSD (n = 90). Speech features were quantified using an automated pipeline for brief recorded samples of free speech. Factor models for the clinical ratings were generated using exploratory factor analysis, then tested with confirmatory factor analysis in the cross-diagnostic and SSD groups. The relationships between factor scores and computational speech features were examined for 202 of the participants.

STUDY RESULTS

We found a 3-factor model with a good fit in the cross-diagnostic group and an acceptable fit for the SSD subsample. The model identifies an impaired expressivity factor and 2 interrelated disorganized factors for inefficient and incoherent speech. Incoherent speech was specific to psychosis groups, while inefficient speech and impaired expressivity showed intermediate effects in people with nonpsychotic disorders. Each of the 3 factors had significant and distinct relationships with speech features, which differed for the cross-diagnostic vs SSD groups.

CONCLUSIONS

We report a cross-diagnostic 3-factor model for speech disturbance which is supported by good statistical measures, intuitive, applicable to SSD, and relatable to linguistic theories. It provides a valuable framework for understanding speech disturbance and appropriate targets for modeling with quantitative speech features.

Machine Learning and Non-Affective Psychosis: Identification, Differential Diagnosis, and Treatment

PURPOSE OF REVIEW

This review will cover the most relevant findings on the use of machine learning (ML) techniques in the field of non-affective psychosis, by summarizing the studies published in the last three years focusing on illness detection and treatment.

RECENT FINDINGS

Multiple ML tools that include mostly supervised approaches such as support vector machine, gradient boosting, and random forest showed promising results by applying these algorithms to various sources of data: socio-demographic information, EEG, language, digital content, blood biomarkers, neuroimaging, and electronic health records. However, the overall performance, in the binary classification case, varied from 0.49, which is to be considered very low (i.e., noise), to over 0.90. These results are fully justified by different factors, some of which may be attributable to the preprocessing of the data, the wide variety of the data, and the a-priori setting of hyperparameters. One of the main limitations of the field is the lack of stratification of results based on biological sex, given that psychosis presents differently in men and women; hence, the necessity to tailor identification tools and data analytic strategies. Timely identification and appropriate treatment are key factors in reducing the consequences of psychotic disorders. In recent years, the emergence of new analytical tools based on artificial intelligence such as supervised ML approaches showed promises as a potential breakthrough in this field. However, ML applications in everyday practice are still in its infancy.

Natural Language Processing and Psychosis: On the Need for Comprehensive Psychometric Evaluation

BACKGROUND AND HYPOTHESIS

Despite decades of "proof of concept" findings supporting the use of Natural Language Processing (NLP) in psychosis research, clinical implementation has been slow. One obstacle reflects the lack of comprehensive psychometric evaluation of these measures. There is overwhelming evidence that criterion and content validity can be achieved for many purposes, particularly using machine learning procedures. However, there has been very little evaluation of test-retest reliability, divergent validity (sufficient to address concerns of a "generalized deficit"), and potential biases from demographics and other individual differences.

STUDY DESIGN

This article highlights these concerns in development of an NLP measure for tracking clinically rated paranoia from video "selfies" recorded from smartphone devices. Patients with schizophrenia or bipolar disorder were recruited and tracked over a week-long epoch. A small NLP-based feature set from 499 language samples were modeled on clinically rated paranoia using regularized regression.

STUDY RESULTS

While test-retest reliability was high, criterion, and convergent/divergent validity were only achieved when considering moderating variables, notably whether a patient was away from home, around strangers, or alone at the time of the recording. Moreover, there were systematic racial and sex biases in the model, in part, reflecting whether patients submitted videos when they were away from home, around strangers, or alone.

CONCLUSIONS

Advancing NLP measures for psychosis will require deliberate consideration of test-retest reliability, divergent validity, systematic biases and the potential role of moderators. In our example, a comprehensive psychometric evaluation revealed clear strengths and weaknesses that can be systematically addressed in future research.

Widespread cortical thinning, excessive glutamate and impaired linguistic functioning in schizophrenia: A cluster analytic approach

Introduction

Symptoms of schizophrenia are closely related to aberrant language comprehension and production. Macroscopic brain changes seen in some patients with schizophrenia are suspected to relate to impaired language production, but this is yet to be reliably characterized. Since heterogeneity in language dysfunctions, as well as brain structure, is suspected in schizophrenia, we aimed to first seek patient subgroups with different neurobiological signatures and then quantify linguistic indices that capture the symptoms of "negative formal thought disorder" (i.e., fluency, cohesion, and complexity of language production).

Methods

Atlas-based cortical thickness values (obtained with a 7T MRI scanner) of 66 patients with first-episode psychosis and 36 healthy controls were analyzed with hierarchical clustering algorithms to produce neuroanatomical subtypes. We then examined the generated subtypes and investigated the quantitative differences in MRS-based glutamate levels [in the dorsal anterior cingulate cortex (dACC)] as well as in three aspects of language production features: fluency, syntactic complexity, and lexical cohesion.

Results

Two neuroanatomical subtypes among patients were observed, one with near-normal cortical thickness patterns while the other with widespread cortical thinning. Compared to the subgroup of patients with relatively normal cortical thickness patterns, the subgroup with widespread cortical thinning was older, with higher glutamate concentration in dACC and produced speech with reduced mean length of T-units (complexity) and lower repeats of content words (lexical cohesion), despite being equally fluent (number of words).

Conclusion

We characterized a patient subgroup with thinner cortex in first-episode psychosis. This subgroup, identifiable through macroscopic changes, is also distinguishable in terms of neurochemistry (frontal glutamate) and language behavior (complexity and cohesion of speech). This study supports the hypothesis that glutamate-mediated cortical thinning may contribute to a phenotype that is detectable using the tools of computational linguistics in schizophrenia.

Progressive changes in descriptive discourse in First Episode Schizophrenia: a longitudinal computational semantics study

Computational semantics, a branch of computational linguistics, involves automated meaning analysis that relies on how words occur together in natural language. This offers a promising tool to study schizophrenia. At present, we do not know if these word-level choices in speech are sensitive to the illness stage (i.e., acute untreated vs. stable established state), track cognitive deficits in major domains (e.g., cognitive control, processing speed) or relate to established dimensions of formal thought disorder. In this study, we collected samples of descriptive discourse in patients experiencing an untreated first episode of schizophrenia and healthy control subjects (246 samples of 1-minute speech; n = 82, FES = 46, HC = 36) and used a co-occurrence based vector embedding of words to quantify semantic similarity in speech. We obtained six-month follow-up data in a subsample (99 speech samples, n = 33, FES = 20, HC = 13). At baseline, semantic similarity was evidently higher in patients compared to healthy individuals, especially when social functioning was impaired; but this was not related to the severity of clinically ascertained thought disorder in patients. Across the study sample, higher semantic similarity at baseline was related to poorer Stroop performance and processing speed. Over time, while semantic similarity was stable in healthy subjects, it increased in patients, especially when they had an increasing burden of negative symptoms. Disruptions in word-level choices made by patients with schizophrenia during short 1-min descriptions are sensitive to interindividual differences in cognitive and social functioning at first presentation and persist over the early course of the illness.

High-Order Language Processing Difficulties in Patients With Schizophrenia: Cross-linguistic and Cross-cultural Results From the Hindi Version of a Newly Developed Language Test

Objective

To detect subtle linguistic performance deficits in patients with schizophrenia, a test battery was developed in Hindi vernacular language.

Method

It was a replication study of observational, analytical, and case-control design. Total of 86 participants, namely 43 patients with schizophrenia and 43 controls, were recruited into the study. The patients were evaluated by using PANSS (positive and negative symptoms scale for schizophrenia) for recruitment into the study. Participants from the general population were evaluated with GHQ-12 (General Health Questionnaire-12) to be found to fit as healthy controls. Subsequently, the linguistic performance of patients (on HLFT: Hindi linguistic function test) was compared with that of controls. The HLFT battery was designed, containing 3 blocks by using antonyms, synonyms, homonyms, hyperonyms, hyponyms, distractors, and adages.

Result

Patients scored significantly less than that of controls in identifying antonyms, distractors, and hyponyms while in identifying homonyms they scored significantly more than that of controls. At block I (antonyms) score of 15.5; at homonym score of 5.5; at hyponym (as in hyponym plus distractor combination) score of 2.5, the sensitivity and specificity for using them as a cutoff to screen for schizophrenia are 60.5% and 67.4%; 86% and 41.9%; 81.4% and 46.5%, respectively.

Conclusion

Ambiguity processing of taxonomic representation such as antonymia, homonymia, hypo-/hyperonymia, synonymia, and also understanding of adages might be significantly impaired in patients with schizophrenia. The HLFT battery could be used as a quick and sensitive instrument to detect and quantify the linguistic difficulties of patients with schizophrenia.

A word is worth a thousand pictures: A 20-year comparative analysis of aberrant abstraction in schizophrenia, affective psychosis, and non-psychotic depression

thinking is a cognitive process that involves the assimilation of concepts reduced from diffuse sensory input, organized, and interpreted in a manner beyond the obvious. There are multiple facets by which abstraction is measured that include semantic, visual-spatial and social comprehension. This study examined the prevalence and course of abstract and concrete responses to semantic proverbs and aberrant abstraction (composite score of semantic, visual-spatial, and social comprehension) over 20 years in 352 participants diagnosed with schizophrenia, affective psychosis, and unipolar non-psychotic depression. We utilized linear models, two-way ANOVA and contrasts to compare groups and change over time. Linear models with Generalized Estimation Equation (GEE) to determine association. Our findings show that regardless of diagnosis, semantic proverb interpretation improves over time. Participants with schizophrenia give more concrete responses to proverbs when compared to affective psychosis and unipolar depressed without psychosis. We also show that the underlying structure of concretism encompasses increased conceptual overinclusion at index hospitalization and idiosyncratic associations at follow-up; whereas, abstract thinking overtime encompasses increased visual-spatial abstraction at index and rich associations with increased social comprehension scores at follow-up. Regardless of diagnosis, premorbid functioning, descriptive characteristics, and IQ were not associated with aberrant abstraction. Delusions are highly and positively related to aberrant abstraction scores, while hallucinations are mildly and positively related to this score. Lastly, our data point to the importance of examining the underlying interconnected structures of 'established' constructs vis-à-vis mixed methods to provide a description of the rich interior world that may not always map onto current quantitative measures.

Natural language processing methods are sensitive to sub-clinical linguistic differences in schizophrenia spectrum disorders

Computerized natural language processing (NLP) allows for objective and sensitive detection of speech disturbance, a hallmark of schizophrenia spectrum disorders (SSD). We explored several methods for characterizing speech changes in SSD (n = 20) compared to healthy control (HC) participants (n = 11) and approached linguistic phenotyping on three levels: individual words, parts-of-speech (POS), and sentence-level coherence. NLP features were compared with a clinical gold standard, the Scale for the Assessment of Thought, Language and Communication (TLC). We utilized Bidirectional Encoder Representations from Transformers (BERT), a state-of-the-art embedding algorithm incorporating bidirectional context. Through the POS approach, we found that SSD used more pronouns but fewer adverbs, adjectives, and determiners (e.g., "the," "a,"). Analysis of individual word usage was notable for more frequent use of first-person singular pronouns among individuals with SSD and first-person plural pronouns among HC. There was a striking increase in incomplete words among SSD. Sentence-level analysis using BERT reflected increased tangentiality among SSD with greater sentence embedding distances. The SSD sample had low speech disturbance on average and there was no difference in group means for TLC scores. However, NLP measures of language disturbance appear to be sensitive to these subclinical differences and showed greater ability to discriminate between HC and SSD than a model based on clinical ratings alone. These intriguing exploratory results from a small sample prompt further inquiry into NLP methods for characterizing language disturbance in SSD and suggest that NLP measures may yield clinically relevant and informative biomarkers.

Baseline Cortical Thickness Reductions in Clinical High Risk for Psychosis: Brain Regions Associated with Conversion to Psychosis Versus Non-Conversion as Assessed at One-Year Follow-Up in the Shanghai-At-Risk-for-Psychosis (SHARP) Study

OBJECTIVE

To assess cortical thickness (CT) and surface area (SA) of frontal, temporal, and parietal brain regions in a large clinical high risk for psychosis (CHR) sample, and to identify cortical brain abnormalities in CHR who convert to psychosis and in the whole CHR sample, compared with the healthy controls (HC).

METHODS

Magnetic resonance imaging, clinical, and cognitive data were acquired at baseline in 92 HC, 130 non-converters, and 22 converters (conversion assessed at 1-year follow-up). CT and SA at baseline were calculated for frontal, temporal, and parietal subregions. Correlations between regions showing group differences and clinical scores and age were also obtained.

RESULTS

CT but not SA was significantly reduced in CHR compared with HC. Two patterns of findings emerged: (1) In converters, CT was significantly reduced relative to non-converters and controls in the banks of superior temporal sulcus, Heschl's gyrus, and pars triangularis and (2) CT in the inferior parietal and supramarginal gyrus, and at trend level in the pars opercularis, fusiform, and middle temporal gyri was significantly reduced in all high-risk individuals compared with HC. Additionally, reduced CT correlated significantly with older age in HC and in non-converters but not in converters.

CONCLUSIONS

These results show for the first time that fronto-temporo-parietal abnormalities characterized all CHR, that is, both converters and non-converters, relative to HC, while CT abnormalities in converters relative to CHR-NC and HC were found in core auditory and language processing regions.

Modeling Incoherent Discourse in Non-Affective Psychosis

BACKGROUND

Computational linguistic methodology allows quantification of speech abnormalities in non-affective psychosis. For this patient group, incoherent speech has long been described as a symptom of formal thought disorder. Our study is an interdisciplinary attempt at developing a model of incoherence in non-affective psychosis, informed by computational linguistic methodology as well as psychiatric research, which both conceptualize incoherence as associative loosening. The primary aim of this pilot study was methodological: to validate the model against clinical data and reduce bias in automated coherence analysis.

METHODS

Speech samples were obtained from patients with a diagnosis of schizophrenia or schizoaffective disorder, who were divided into two groups of n = 20 subjects each, based on different clinical ratings of positive formal thought disorder, and n = 20 healthy control subjects.

RESULTS

Coherence metrics that were automatically derived from interview transcripts significantly predicted clinical ratings of thought disorder. Significant results from multinomial regression analysis revealed that group membership (controls vs. patients with vs. without formal thought disorder) could be predicted based on automated coherence analysis when bias was considered. Further improvement of the regression model was reached by including variables that psychiatric research has shown to inform clinical diagnostics of positive formal thought disorder.

CONCLUSIONS

Automated coherence analysis may capture different features of incoherent speech than clinical ratings of formal thought disorder. Models of incoherence in non-affective psychosis should include automatically derived coherence metrics as well as lexical and syntactic features that influence the comprehensibility of speech.