Abnormal eye movements in psychotic patients were first described more than 100 years ago.(Diefendorf AR, Dodge R (1908): An experimental study of the ocular reactions of the insane from photographic records. Brain 31:451– 489). Over the years several attempts have been made to explore this defect as a marker of the disease. Schizophrenic patents and family members have deficits in smooth pursuit eye movements. Incidence vary between 12-95%. However, up to 20% controls also show this defect. Gaze maintenance is also defective in schizophrenia, though not all studies agree on this.Free moving scan paths, where subjects view pictures (free-viewing) show that eye movements are confined to limited area/ limited features of the source.
This group of researchers ( Aberdeen and Munich Universities) studied the above defects in cases and controls and tried various classifier models to find the best model to distinguish between cases and controls.
88 schizophrenia patients, mostly outpatients on second generation antipsychotics, participated in the study. Any neurological condition/head injury was an exclusion criteria.Controls were free of substance abuse and family history of psychosis.Infrared eye-movement recording was performed using head- mounted equipments with a remote camera and head/chin rest.Smooth pursuit involved tracking a circular target for 20 sec as it moved . Free-viewing scanpaths were produced in response to 56 color images (8-sec each). Gaze maintenance was tested (Fixation stability test) by maintaining steady gaze only on a central circular target for 5 sec and to ignore an identical distracter target.
Case control data was used to generate models. This included gradient boosted decision trees (GBDT) and probabilistic neural networks (PNN). The model used all 55 dependent measures in the above tests.
Performance on pursuit, scene viewing, and steady fixation tasks were all abnormal in the schizophrenia group compared with control subjects
The boosted decision tree (GBDT) method achieves close to complete separation on single tasks. Using combined measures from all tasks GBDT achieves complete separation of the training data and cross- validated performance is good (accuracy 87%).
Retesting (stability): The GBDT model was used to check for a change in predicted status of 26 schizophrenia and 8 control individuals attending for reassessment 9 months later, with 82.4% accuracy.All control subjects were correctly classified.
Novel cases: Predictive performance was tested using novel schizophrenia (n= 36) and control volunteers (n =13) recruited at a later date. Six schizophrenia cases and two control subjects were misclassified (79.6% accuracy).
Modified PNN reached 98% accuracy in discriminating cases and controls. A simplified PNN model using only the free viewing and fixation tasks produced accuracy of 95%.
The group consider that the free- viewing scan paths represent a generalsed deficit in schizophrenia. This appears to be the biggest discriminator of cases and controls.
It also appear that these abnormalities are state markers.(abnormalities are independent of effect of medications, independent of duration of illness, age of onset or severity of illness).
Some differences in German and Scottish samples.High proportion of cases smoked.Replication cohort was of much younger age. Authors discuss the reasons for these and how these factors may not affect the observations
1. Is this specific to schizophrenia? further research required with other disorder groups
2. How does this fit with our current understanding of patho physiology of schizophrenia?
Clinical practice: Eye movement recordings are less time consuming and easy to roll out. Replication in further samples is eagerly awaited.
Summary of the article:
Simple viewing tests can detect eye movement abnormalities that distinguish schizophrenia cases from controls with exceptional accuracy. Benson PJ, Beedie SA, Shephard E, Giegling I, Rujescu D, St Clair D.Biol Psychiatry. 2012 Nov 1;72(9):716-24.