A multicentric study examining a deep-learning-based computer model for classifying bipolar disorder using retinal...

Area of Science:

• Ophthalmology and Psychiatry
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• The retina serves as a "window to the brain," with vascular abnormalities linked to bipolar disorder (BD).
• Deep learning (DL) offers advanced computational methods for analyzing medical images, including retinal vasculature.
• Previous research has not utilized DL for classifying bipolar disorder using retinal vascular images.

Purpose of the Study:

• To investigate the efficacy of a deep learning model in classifying individuals with bipolar disorder (BD) versus healthy volunteers (HV) using retinal fundus images.
• To assess the transfer learning capability of the developed model on an independent test dataset.

Main Methods:

• A cohort of 383 participants (188 BD, 195 HV) was studied, with data split into training (327) and testing (56) sets.
• Retinal fundus images were captured using a non-mydriatic fundus camera.
• An optimized convolutional neural network (CNN) model was trained and its transfer learning performance evaluated.

Main Results:

• The CNN model demonstrated high performance on the training dataset: 88.0% sensitivity, 85.7% specificity.
• On the independent test dataset, the model achieved 85.7% accuracy, 83.3% sensitivity, and 92.9% specificity.
• The balanced model parameters and strong performance on the test set indicate robustness.

Conclusions:

• The developed CNN model effectively differentiates between bipolar disorder patients and healthy individuals.
• The study highlights the potential clinical utility of deep learning analysis of retinal images for psychiatric diagnostics.
• Successful replication in an independent test dataset supports the model's transfer learning capabilities.