Characterizing functional regional homogeneity (ReHo) as a B-SNIP psychosis biomarker using traditional and machine learning approaches

Lanxin Ji, Shashwath A. Meda, Carol A. Tamminga, Brett A. Clementz, Matcheri S. Keshavan, John A Sweeney, Elliot S. Gershon, Godfrey D. Pearlson

Research output: Contribution to journalArticle

Abstract

Background: Recently, a biologically-driven psychosis classification (B-SNIP Biotypes) was derived using brain-based cognitive and electrophysiological markers. Here, we characterized a local functional-connectivity measure, regional homogeneity (ReHo), as a biomarker across Biotypes and conventional DSM diagnoses. Methods: Whole-brain ReHo measures of resting-state functional MRI were examined in psychosis patients and healthy controls organized by Biotype and by DSM-IV-TR diagnosis (n = 737). Group-level ANOVA and individual-level prediction models using support vector machines (SVM) were employed to evaluate the discriminative characteristics in comparisons of 1) DSM diagnostic groups, 2) Biotypes, to controls, and 3) within-proband subgroups with each other. Results: Probands grouped by Biotype versus controls showed a unique abnormality pattern: Biotype-1 displayed bidirectional ReHo differences in more widespread areas, with higher ReHo in para-hippocampus, fusiform, inferior temporal, cerebellum, thalamus and caudate, plus lower ReHo in the postcentral gyrus, middle temporal, cuneus, and middle occipital cortex; Biotype-2 and Biotype-3 showed lesser and unidirectional ReHo changes. Among diagnostic groups, only schizophrenia showed higher ReHo versus control values in the inferior/middle temporal area and fusiform gyrus. For within-patient comparisons, Biotype-1 showed characteristic ReHo when compared to Biotype-2 and Biotype-3. SVM results more accurately identified Biotypes than DSM diagnoses. Conclusion: We characterized patterns of ReHo abnormalities across both Biotypes and DSM sub-groups. Both group-level statistical and machine-learning methods were more sensitive in capturing ReHo deficits in Biotypes than DSM. Overall ReHo is a robust psychosis biomarker.

Original languageEnglish (US)
JournalSchizophrenia Research
DOIs
StateAccepted/In press - Jan 1 2019

    Fingerprint

Keywords

  • Biological marker
  • Biotypes
  • Machine learning
  • Psychosis
  • ReHo
  • Resting-state

ASJC Scopus subject areas

  • Psychiatry and Mental health
  • Biological Psychiatry

Cite this