Optimized Smartphone-based Implementation of B-COSFIRE Filter for Retinal Blood Vessel Segmentation

Abstract

Medical professionals widely perform retinal blood vessel analysis from fundus images to detect and diagnose various ocular and systemic diseases. Many studies have proposed methods for automatic blood vessel extraction from fundus images which are mostly desktop-based implementations for analysing images from table-top fundus cameras. In resource-limited settings such as clinics in rural or remote areas, desktop fundus cameras are rarely available and workstations for automatic retinal analysis can be difficult to obtain and set up. In these recent years, handheld fundus cameras have been developed to offer a mobile solution for retinal screening at a cheaper cost. However, research on mobile processing for retinal analysis of these handheld fundus images is still limited. We propose an optimized retinal blood vessel segmentation method for an Android-based smartphone platform using the Bar-Combination of Shifted Filter Response (B-COSFIRE) filter. Since there is no public database for handheld fundus images to date, the developed mobile retinal blood vessel segmentation application was evaluated using two publicly available table-top fundus images databases, DRIVE and STARE, with achieved segmentation accuracy of 94.87% and 95.96% respectively. The results showed that the method not only achieved comparable performance to published methods but is also faster, making it a possible cost-effective and efficient option for retinal diagnosis in rural areas.