AI Solutions

DEEP:FUNDUS-DR-01

AI Solution: DEEP:FUNDUS-DR-01
Core Technology: Funduscopy image processing analysis and diabetic retinopathy analysis technology

Diabetic retinopathy is a disease in which patients with diabetes can damage blood vessels in the retina, leading to loss of vision or blindness. When diabetic retinopathy progresses, bleeding occurs from blood vessels in the retina and is injected into the vitreous body. From this phenomenon, black spots or streaks appear in the eye.

Overview

Our product is a product that classifies the degree of suspected diabetic retinopathy in Funduscopy images using artificial intelligence technology. Through this, based on information on the classification of diabetic retinopathy, it can be helpful in diagnosis and treatment of medical personnel.

AI Core Technology

This technology is an algorithm that categorizes and informs the severity of the input fundus image, and shows the result in 5 levels of severity, from images without diabetic retinopathy (0) to diffuse disease (4). Resize to unify the size of the input image to a specific size, Contrast-Limited Adaptive Histogram Equalization (CLAHE) to emphasize the blood vessels of the fundus image by adjusting the histogram to emphasize the light and dark areas, and to set the 8-bit image to a value of 0~1. Pre-processing techniques such as maximum-minimum regularization were used to normalize and increase the stability of artificial intelligence model training.

The deep learning model was optimized using the ResNet-based model, and through the efficient operation method of the Residual Block structure, we were able to handle high-resolution endoscopic images and obtain good performance.