Automated Schizophrenia Detection and Classification (STFDL-ASDC): A Proposed Model for Self-Reported Schizophrenic Episodes
Jayashree M Kudari, Dyuti Ganesh · 2025
Schizophrenia is a severe mental illness that disrupts brain functions such as perception and thought, leading to profound impacts on individuals’ lives. Its early diagnosis is complex, often hindered by multiple comorbidities that challenge effective patient management and reduce the likelihood of positive outcomes. This study aims to enhance the detection of schizophrenic episodes through a novel deep learning-based Automated Schizophrenia Detection and Classification (STFDL-ASDC) model that uses self-recorded video data. Our approach extracts spatial and temporal features from the videos through a Two-Stream Inflated 3D ConvNet model. The spatial stream processes facial expressions, body language, and gestures, while the temporal stream analyzes the progression of emotional and behavioral patterns. To improve video quality, preprocessing techniques like Wiener filtering and adaptive histogram equalization (AHE) are applied. Spatial features are extracted from the processed frames using the RegNetY002 CNN architecture, with temporal feature analysis and classification managed by the Inflated 3D-ConvNet. This research highlights the proposed model’s ability to provide real-time observation of schizophrenia symptoms, potentially enabling early intervention and tailored treatment strategies.