O11.2. CHARACTERIZING CANNABINOID INDUCED ACUTE PERSISTENT PSYCHOSIS (CIAPP) AS A POSSIBLE SUBTYPE OF SCHIZOPHRENIA USING DEEP LEARNING

The schizophrenia syndrome likely encompasses multiple clinically related illness manifestations that result from distinct etio-pathological processes converging onto fewer final common pathways. Exposure to cannabis is known to result in a syndrome that clinically mimics schizophrenia-psychosis, outlasts the acute intoxication, persists for days to weeks, requires clinical intervention, and may recur with cannabis exposure. Characterizing the vulnerability to Cannabinoid Induced Acute and Persistent Psychosis (CIAPP), its clinical and neuro-physiological correlates, and its relationship to schizophrenia may enhance our understanding of the neurobiology of schizophrenia in general and specifically, this subtype. Deep learning is an extremely powerful approach to classify (e.g., cancerous vs healthy cells) and use complex stimuli to anticipate future outcomes (e.g., hurricane path) with high accuracy. Thus, deep learning seems a promising approach to differentiate subtypes of complex syndromes such as schizophrenia. In a prospective case-control study at Central Institute of Psychiatry, Ranchi, India, we compared hospitalized cases of CIAPP with two control groups which included: 1) hospitalized cases with psychosis unrelated to cannabis, and (PUC) 2) healthy controls (HC). Demographic and substance use variables, and familial loading for psychiatric illnesses (FIGS) were evaluated at baseline. The following assessments were carried out at four time points - baseline, mid hospitalization, at discharge, and at 6 months post discharge: 1) measures of psychosis (PANSS), mood (YMRS and Calgary depression scale) and cognition (Cogstate battery); 2) psychophysiological variables including resting and Auditory Steady State Response (ASSR) EEG. Electrophysiological and behavioral data were integrated into subject-specific neurobehavioral “fingerprints”, i.e. graph-like objects depicting EEG (2 second epochs) and behavioral information. These fingerprints (~200 per subject) were used to train the classification apparatus of a deep convolutional network (Inception-Res v2) pre-trained for image classification. The trained network was tested on a validation data set from an independent sub-sample of subjects. Data has been collected for 50 consecutive CIAPP cases and 25 controls (15 PUC, 10 HC) and a part of the sample has completed the sixth month follow up. Interim analysis of the data from baseline, mid-hospitalization and at discharge time points suggests that in comparison to PUC, CIAPP has a distinct profile with equivalent psychosis symptoms but more mania-like symptoms and lower pre-morbid schizotypy scores. They also have lower scores on cognitive tests at baseline in specific neurocognitive domains including working memory and recall tests but had better performance in paired associate learning and social cognition tests. Electrophysiological data showed that CIAPP and PUC had reduced gamma-band neural connectivity compared to HC while CIAPP showed levels of gamma-band power comparable to HC. Post-training, the neural network identified and classified neurobehavioral fingerprints from CIAPP, PUC, and HC with >98% accuracy in both the training and independent validation data sets. The preliminary results of this investigation suggest that CIAPP represents a subtype of schizophrenia with distinct neuro-behavioral correlates. Furthermore, deep learning has shown to be useful to classify such disease subtypes. Using this approach in larger training and validation data sets, and inclusion of the longitudinal data from 6-month follow-up may improve the robustness of the neural net classifier.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.