A Multi-Label Computer-aided Diagnoses System for Detecting and Diagnosing Diabetic Retinopathy

Abstract

Multi-label classification (MLC) is considered an active research topic, especially in medical image analysis. We used MLC to diagnose the multiple different grades of Diabetic Retinopathy (DR). DR is caused when a patient's blood pressure and blood sugar are too high and results in damage in the blood vessels (BVs). BVs supply the blood to the retina. If the retina does not get the blood it needs, it can eventually cause permanent blindness. The early diagnosis of different DR grades leads the ophthalmologists to efficient treatment. In this paper, we developed a multi-label computer-aided diagnosis (ML-CAD) system to apply MLC for different DR grades using color fundus images. Our system utilizes 11 texture features descriptors by retrieving the average of the Gray Level Run Length Matrix (GLRM) on four directions 0°, 45°, 90°, and 135°. It distinguishes the normal from DR cases by supplying the extracted features to the support vector machine (SVM) classifier. Then, the proposed CAD system segments some significant characteristics from DR fundus images, which are BV, exudates (EX), microaneurysms (MA), and hemorrhages (HM). After that, it calculates the Gray Level Co-occurrence Matrix (GLCM), regions of interest (ROIs) areas, and BV bifurcation point's calculations. Finally, the feature vector is trained and tested using a multi-label SVM (MSVM) classifier generates a suitable DR grade. We used four various benchmark datasets to evaluate the performance of our system in terms of accuracy (ACC), sensitivity (SEN), specificity (SPE), the area under the curve (AVC), and micro F1 measure. The experiments confirmed that our ML-CAD system outperforms the other diagnosing DR systems.