Multimodal and multi-omics-based deep learning model for screening of optic neuropathy

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optic nerve diseases, including diabetic optic neuropathy, glaucomatous optic neuropathy, and optic neuritis, pose significant diagnostic challenges.
• Current screening methods may lack the precision needed for early and accurate detection.

Purpose of the Study:

• To evaluate the efficacy of multimodal data and multi-omics strategies in conjunction with deep learning for optic nerve disease screening.
• To compare the diagnostic performance of different deep learning models utilizing various data modalities.

Main Methods:

• A retrospective study involving 156 patients with optic nerve diseases.
• Development of a deep learning model using fundus photography and infrared reflectance (IR) images.
• Analysis of data using traditional omics, Resnet101, and fusion models in single and multimodal settings.

Main Results:

• Multimodal data integration improved accuracy, with fusion models achieving 0.99 accuracy on the training set and 0.88 on the test set.
• Macro-average AUCs for multimodal data reached 0.96 (traditional omics), 0.97 (Resnet101), and 0.99 (fusion model).
• The fusion model demonstrated superior performance in distinguishing between different types of optic neuropathies.

Conclusions:

• Deep learning models leveraging multimodal data and multi-omics strategies significantly enhance the accuracy of screening and diagnosis for major optic nerve diseases.
• This integrated approach holds promise for improving the early detection and management of optic neuropathies.