Diabetic Macular Edema Optical Coherence Tomography Biomarkers Detected with EfficientNetV2B1 and ConvNeXt

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetes mellitus (DM) poses a significant healthcare challenge, necessitating efficient screening for vision-threatening complications like diabetic macular edema (DME).
• Current screening methods face limitations in time and accessibility, highlighting the need for advanced diagnostic tools.
• Optical coherence tomography (OCT) is crucial for DME detection, but analyzing large datasets requires efficient methods.

Purpose of the Study:

• To evaluate the efficacy of ConvNeXt and EfficientNet AI architectures in identifying DME from OCT images.
• To assess the potential of AI in increasing screening capacity and speed for diabetic retinopathy (DR) complications.
• To determine if AI models pretrained on natural images can effectively analyze medical OCT scans.

Main Methods:

• Utilized ConvNeXt and EfficientNet architectures for analyzing real-world OCT images.
• Developed models to differentiate between healthy scans and those with diabetic retinopathy (DR).
• Trained a model to detect edema, retinal detachment, and hyperreflective foci without pixel-level annotation.

Main Results:

• Achieved 0.98 accuracy in differentiating between diabetic retinopathy (DR) and healthy retinal scans.
• Successfully developed a model capable of indicating the presence of DME biomarkers (edema, detachment, hyperreflective foci) without pixel-level annotation.
• Demonstrated that pretrained networks (ConvNeXt, EfficientNet) effectively identify relevant features in OCT scans, even when trained on natural images.

Conclusions:

• AI models like ConvNeXt and EfficientNet can accurately identify features for differentiating healthy retinas from DR, leveraging knowledge from natural image pretraining.
• AI enables the detection and localization of DME biomarkers without the need for detailed pixel-level annotations.
• AI-assisted diagnosis of DME shows promise in reducing healthcare costs, enhancing patient quality of life, and decreasing wait times for ophthalmological consultations and treatment.