BACKGROUND: We have investigated which eye-movement tests alone and combined can best discriminate schizophrenia cases from control subjects and their predictive validity. METHODS: A training set of 88 schizophrenia cases and 88 controls had a range of eye movements recorded; the predictive validity of the tests was then examined on eye-movement data from 34 9-month retest cases and controls, and from 36 novel schizophrenia cases and 52 control subjects. Eye movements were recorded during smooth pursuit, fixation stability, and free-viewing tasks. Group differences on performance measures were examined by univariate and multivariate analyses. Model fitting was used to compare regression, boosted tree, and probabilistic neural network approaches. RESULTS: As a group, schizophrenia cases differed from control subjects on almost all eye-movement tests, including horizontal and Lissajous pursuit, visual scanpath, and fixation stability; fixation dispersal during free viewing was the best single discriminator. Effects were stable over time, and independent of sex, medication, or cigarette smoking. A boosted tree model achieved perfect separation of the 88 training cases from 88 control subjects; its predictive validity on retest assessments and novel cases and control subjects was 87.8%. However, when we examined the whole data set of 298 assessments, a cross-validated probabilistic neural network model was superior and could discriminate all cases from controls with near perfect accuracy at 98.3%. CONCLUSIONS: Simple viewing patterns can detect eye-movement abnormalities that can discriminate schizophrenia cases from control subjects with exceptional accuracy.
BACKGROUND: We have investigated which eye-movement tests alone and combined can best discriminate schizophrenia cases from control subjects and their predictive validity. METHODS: A training set of 88 schizophrenia cases and 88 controls had a range of eye movements recorded; the predictive validity of the tests was then examined on eye-movement data from 34 9-month retest cases and controls, and from 36 novel schizophrenia cases and 52 control subjects. Eye movements were recorded during smooth pursuit, fixation stability, and free-viewing tasks. Group differences on performance measures were examined by univariate and multivariate analyses. Model fitting was used to compare regression, boosted tree, and probabilistic neural network approaches. RESULTS: As a group, schizophrenia cases differed from control subjects on almost all eye-movement tests, including horizontal and Lissajous pursuit, visual scanpath, and fixation stability; fixation dispersal during free viewing was the best single discriminator. Effects were stable over time, and independent of sex, medication, or cigarette smoking. A boosted tree model achieved perfect separation of the 88 training cases from 88 control subjects; its predictive validity on retest assessments and novel cases and control subjects was 87.8%. However, when we examined the whole data set of 298 assessments, a cross-validated probabilistic neural network model was superior and could discriminate all cases from controls with near perfect accuracy at 98.3%. CONCLUSIONS: Simple viewing patterns can detect eye-movement abnormalities that can discriminate schizophrenia cases from control subjects with exceptional accuracy.
Authors: Arnim Johannes Gaebler; Klaus Mathiak; Jan Willem Koten; Andrea Anna König; Yury Koush; David Weyer; Conny Depner; Simeon Matentzoglu; James Christopher Edgar; Klaus Willmes; Mikhail Zvyagintsev Journal: Brain Date: 2015-03-04 Impact factor: 13.501