Poster Session F, Tuesday, March 26, 8:00 – 10:00 am, Pacific Concourse
Identification of first-episode psychosis from the brain activity of subjects viewing a naturalistic stimulus: Time-window-based neural network analysis
Vesa Vahermaa1, Athanasios Gotsopoulos1, Jussi Alho1, Mikko Sams1, Tuukka Raij1; 1Aalto University, School of Science
The neuronal basis of psychosis is not fully understood and diagnosis primarily entails non-objective methods such as interviews and questionnaires. Modern machine learning methods are able to identify patterns from large amounts of data, including functional magnetic resonance imaging (fMRI) data. Here, we presented a shortened movie (Alice in Wonderland, by Tim Burton, 2010) to our subjects with diagnosed first-episode psychosis (N=51) and controls (N=33) during whole-brain fMRI (voxel size 2mm, TR 1.8s). (54 male, average age of 25.14 years (SD=5.46)) We applied a neural network classifier to predict whether a subject belonged to the patient vs. control group. We applied this analysis to nine-second sliding windows. The classification accuracy throughout the movie was close to chance level (from 42.31% to 62.56%, mean 52.10%, SD=5.73%), with certain time points exhibiting significantly above chance level accuracy. While our results do not convey the statistical power to be used for clinical purposes, they serve as a starting point for future research on diagnostic tools based on naturalistic stimuli.
Topic Area: METHODS: Neuroimaging