Diabetic eye sentinel: prescreening of diabetic retinopathy using retinal images obtained by a mobile phone camera

Currently, there are few ophthalmologists in Thailand and other developing countries, and their number has increased slowly. However, there is a rapid growth of patients suffering from diabetes mellitus. This makes conventional diabetic retinopathy (DR) prescreening a difficult task. Advances in digital technologies allow the use of portable handheld cameras connected to a mobile phone to obtain retinal fundus images. However, the quality of the mobile phone images is significantly lower and the field of view is narrower than the images from the standard slit-lamp machines. This makes the existing segmentation techniques inefficient. The purpose of this work is to verify the applicability of a hand-held camera with fast computational methods integrated into a mobile phone. Therefore, the paper offers an effective solution called the diabetic eye sentinel (DES). The system segments hemorrhages and exudates in the mobile phone fundus images and detects the DR. The DES is composed of a front-end user interface, back-end computational modules based on a decision tree, the HSV color model, and the edge map. We discuss and analyze the accuracy of the system applied to a dataset of mobile phone images (size = 134) from patients diagnosed with diabetes mellitus at the Thammasat Chalermprakiat Hospital, Pathum Thani, Thailand. The results are compared to a previous version of the algorithm that was based on adaptive thresholding and two other conventional segmentation methods. The numerical experiments show that the proposed system achieves an acceptable accuracy (95.23%), a low false-negative rate (4.76%), and good speed. The proposed system can be used under real clinical conditions.

Automated summary

The study aims to verify the feasibility of integrating a handheld camera with fast computational methods into a mobile phone, proposing an effective solution named DES for segmenting abnormalities in mobile phone fundus images and detecting diabetic retinopathy.