Abstract

Background: Brain disorders are conditions that affect brain structure, function, or chemistry, causing various symptoms and impairments. Brain disorders are categorized into neurodegenerative disorders, such as Alzheimer’s disease and Parkinson’s disease, which involve progressive neuron degeneration; mental health disorders, including depression, anxiety, bipolar disorder, and schizophrenia; and traumatic brain injuries caused by external force resulting in temporary or permanent brain damage. Mood disorders, including major depressive disorder (MDD) and bipolar disorder (BD), are frequently underdiagnosed, contributing to significant clinical burden. To address this challenge, we introduce a novel computational framework that leverages multimodal data integration by combining patient-specific structural magnetic resonance imaging (sMRI) with whole-exome sequencing (WES) data.
Methods: Our dataset consisted of brain imaging and genetic data from 321 East Asian individuals, including 147 diagnosed with major depressive disorder (MDD), 78 with bipolar disorder (BD), and 96 healthy controls. In addition, we used child brain magnetic resonance imaging (MRI) for external validation.
We initially prepared and preprocessed our data for the adult dataset. We then loaded SNP data from a CSV file, normalized the features, and preprocessed MRI images stored in NIFTI format by resizing and augmenting them. Various deep learning models (e.g., InceptionV3, ResNet) were then employed to extract features from MRI data. SNP characteristics were extracted from the preprocessed genetic data. We aligned the number of samples in the SNP and MRI feature sets, concatenated these features to form a combined feature set, and normalized the combined features.
Results: Combined features are input into machine learning classifiers (e.g., Support Vector Machine [SVM], K Nearest Neighbours) for final classification with the best accuracy of 74.2% on a linear SVM classifier for detection of mood disorders. Further, we considered two more results, with the second being the classification of the child brain MRI dataset into abnormal and normal categories, achieving an exceptional accuracy of 99.8% on the cubic SVM classifier.
Conclusion: Our approach can support the diagnostic evaluation of psychiatric patients by providing the incorporation of additional neuroimaging modalities and genomic information into routine clinical workflows.