Multi-modal deep learning model for bipolar depression adolescents with verbal auditory hallucinations

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Adolescent bipolar depression (ABD) with verbal auditory hallucinations (AVHs) presents diagnostic challenges.
• Accurate classification is crucial for timely and effective treatment.

Purpose of the Study:

• To develop and evaluate a multimodal deep learning model for classifying adolescent bipolar depression with AVHs.
• To integrate clinical data and neuroimaging features for improved diagnostic accuracy.

Main Methods:

• A retrospective analysis of 47 untreated adolescent bipolar depression patients.
• Collection of clinical data (sex, age, education, self-harm behaviors) and 1H-MRS scans of the ventromedial prefrontal cortex (vmPFC).
• Construction of a multimodal deep learning model using clinical and MRS-derived features.

Main Results:

• The best-performing model achieved a classification accuracy of 71.43% on a fixed test set.
• The model demonstrated balanced performance with precision, recall, and F1-score all reaching 0.75.
• Preliminary technical feasibility was indicated for the multimodal approach on a small dataset.

Conclusions:

• A novel multimodal Transformer-based framework was proposed for classifying adolescent bipolar depression with AVHs.
• The model effectively integrates heterogeneous data, including clinical and neuroimaging features.
• The findings suggest the potential of advanced deep learning architectures for diagnosing complex psychiatric conditions.