Automated Bipolar Disorder Detection and Phenotype Stratification via Deep Learning Models

Accurate diagnosis and phenotype stratification are difficult for bipolar disorder (BD), a complex neuropsychiatric disorder influenced by both genetic and environmental factors. Although deep learning-based predictive modeling is still underutilized, Structural MRI (sMRI) and functional MRI (fMRI) with Electroencephalography (EEG) recordings provides important insights into brain abnormalities linked to BD. Using,this paper suggests a hybrid deep learning framework for the automated detection and phenotypic classification of BD. In order to capture spatial and temporal brain patterns, the preprocessing pipeline consists of image standardization, time-series extraction, and feature selection to preserve the most informative elements. These are then converted into NIFTI objects. The system models temporal and spatial dependencies by combining long short-term memory (LSTM) networks with artificial neural networks (ANNs). According to experimental results, the suggested framework outperforms traditional machine learning techniques, achieving a 96% predictive accuracy. These results demonstrate the potential of deep learning models for clinical decision support and precision diagnostics in BD, enabling accurate phenotypic variation identification and stratification.