A Multitask Deep-Learning System to Classify Diabetic Macular Edema for Different Optical Coherence Tomography...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic macular edema (DME) is a leading cause of vision loss in diabetes mellitus (DM).
• Accurate and timely diagnosis of DME is crucial for effective management.
• Current diagnostic methods may benefit from advanced automated tools.

Purpose of the Study:

• To develop and validate a deep learning (DL) system for classifying DME.
• To assess the system's performance across different optical coherence tomography (OCT) devices.
• To evaluate the DL system's potential as a screening tool for DME.

Main Methods:

• A multitask convolutional neural network (CNN) using ResNet backbone was trained on 73,746 OCT images.
• Two versions of the CNN were developed: 3D volume scans and 2D B-scans.
• External validation was performed on 26,981 images from seven independent datasets across multiple countries.

Main Results:

• The DL system achieved high performance in classifying DME, with area under the receiver operating characteristic curves (AUROCs) ranging from 0.937 to 0.965 on the primary dataset.
• AUROCs exceeded 0.906 for external datasets, demonstrating robust generalization.
• Classification of DME subgroups (center-involved DME vs. non-center-involved DME) also yielded high AUROCs (>0.894).

Conclusions:

• The developed DL system demonstrates excellent performance in automated DME classification.
• This AI tool holds significant potential as a second-line screening tool for diabetic patients.
• Implementation could lead to more efficient triaging of patients to eye clinics.