Lesion Detection

Early signs of any retinal diseases are usually reflected in retinal lesions such as hemorrhages, microaneurysms or exudates in fundus photographs. The detection of these lesions is carried out manually on a routine basis by most ophthalmologists, which can lead to intensive effort due to the size of the lesions and low-quality images.

A2DS team aims to tackle this problem with our new feature: lesion detection. The new feature highlights the region under suspicion of hemorrhages, microaneurysms, and exudates, allowing ophthalmologists to integrate quintessential findings from the image and predicted result of diseases from our system before coming to a definitive diagnostic decision. Interesting enough, the disease progress can also be monitored with the quantification of lesions within the image. In one of our discussion with ophthalmologists, this new feature is expected to aid many ophthalmologists in further analysis of disease and improve the overall diagnostic procedure.

‍

Our lesion detection algorithm is based on one of the state-of-the-art Deep Learning techniques in segmentation and has proved effectively with high precision of detection. In some cases, the AI-powered system is able to quantify small-sized lesions which are often missed by human eyes.

The detection of such lesions is meaningful in discovering early stages of the disease, enabling in-time support from ophthalmologists before the disease starts worsening.