Correlation of Audiovisual Features With Clinical Variables and Neurocognitive Functions in Bipolar Disorder, Mania

Istanbul Saglik Bilimleri University89 enrolled

Overview

The aim of this study is to show the physiological changes during manic episode in bipolar mania how much they differentiate from remission and healthy control. Relation of audio-visual features as physiological changes and cognitive functions and clinical variables will be searched. The aim is to find biologic markers for predictors of treatment response via machine learning techniques to be able to reduce treatment resistance and give an idea for personalized treatment of bipolar patients.

The objective of this research protocol is to find audio-visual features which differentiates bipolar mani/ remission/ health/ simulation and predicts treatment response earlier and detect neurocognitive changes during mania/ remission and difference from the healthy control. During hospitalization in every follow up day (0th- 3rd- 7th- 14th- 28th day) and after discharge on the 3rd month, presence of depressive and manic features for patients was evaluated using Young Mania Rating Scale(YMRS) and Montgamery- Asberg Depresyon Scale (MADRS). Audiovisual recording is done by a video camera in every follow up day for patients and for healthy controls which includes also depression and mania simulation. Cambridge Neurophysiological Assessment Battery (CANTAB) were administered to both groups( for patients both in the manic phase and in the remission) to assess neurocognitive functions.

Study Type

OBSERVATIONAL

Enrollment

Conditions

Bipolar Disorder, Manic Artificial Intelligence Neurocognition Treatment Resistant Disorders

Interventions

Ongoing treatment for bipolar maniaDRUG

Prescribed by the following doctor during hospitalization and after discharge

Audiovisual recording during guided presentationDIAGNOSTIC_TEST

Seven tasks such as explaining the reason to come to hospital/participate in the activity, describing happy and sad memories, counting up to thirty, explaining two emotion eliciting pictures

Eligibility

Sex: ALLMin age: 18 YearsMax age: 65 YearsHealthy volunteers:

Medical Language ↔ Plain English

Inclusion Criteria: * diagnosis of BD type I, manic episode according to DSM-5 \[10\] given by the following doctor, * being informed of the purpose of the study and having given signed consent before enrollment. Exclusion Criteria: * being younger than 18 years or older than 60 years, * showing low mental capacity during the interview * expression of hallucinations and disruptive behaviors during the interview, * presence of severe organic disease, * presence of any organic disease that may affect cognition * having less than five years of public education * diagnosis of substance or alcohol abuse in the last three months (except nicotine and caffeine) * presence of cerebrovascular disorder, head trauma with longer duration of loss of consciousness, severe hemorrhage and dementia, * having electroconvulsive therapy in the last one year. For the healthy control group, the following additional criteria were considered for exclusion * presence of family history of mood or psychotic disorder, * presence of psychiatric disorder during interview or in the past.

Outcomes

Primary Outcomes

Treatment response

The proportion of Young Mania Rating Scale(YMRS) score ( at baseline to 3rd- 7th- 14th- 28th day and 3rd month ( Baseline scale/ Follow-up day scale) YMRS score utilized rating scales to assess manic symptoms ranged between 0-76 1. Remission: Yt \<= 7 2. Hypomania: 7 \< Yt \< 20 3. Mania: Yt \>= 20.

Time frame: from baseline until 3rd month

Changes in visual features

Functionals of appearance descriptors extracted from fine-tuned Deep Convolutional Neural Networks (DCNN), geometric features obtained using tracked facial landmarks (Unweighted Average Recall) Geometric frame level 23 geometric features and apperance descriptors 4096 dimensional features from the last convolutional layer of the FER fine-tuned CNN which are summarized via mean and range functionals over sub-clips and the decisions are voted at video level, an UAR performance is obtained. Feature vectors extracted from video is modelled using Partial Least Squares (PLS) regression and Extreme Learning Machines classifiers Unweighted Average Recall (UAR), which is mean of class-wise recall scores, is commonly used as performance measure, instead of accuracy, which can be misleading in the case of class-imbalance