Diabetic retinopathy detection and grading of retinal fundus images using coyote optimization algorithm with deep learning

Diabetic retinopathy (DR) is a major reason of preventable blindness for diabetic patients. Regular retinal screening is recommended for diabetic persons to detect DR at the earlier stages. Manual retinal screening of DR is a difficult and laborious process, computer aided diagnosis models become essential. Recently deep learning (DL) methods enable effectual detection and classification of medical images, particularly retinal fundus images. With this motivation, this work presents an intelligent coyote optimization algorithm with DL based DR detection and grading (ICOA-DLDRD) model on retinal fundus images. The purpose of the ICOA-DLDRD approach is to identify the presence of DR on retinal fundus images. Primarily, the ICOA-DLDRD algorithm comprises Gabor filtering (GF) based noise removal and optimal region growing segmentation technique. Further, the primary seed points and thresholds of the region growing segmentation technique are optimally created utilizing the glowworm swarm optimization (GSO) algorithm. In addition, SqueezeNet with class attention learning (CAL) layer is derived for the extraction of feature vectors. Lastly, COA with a deep extreme learning machine (DELM) classifier is applied for the detection and grading of DR, in which the penalty parameter C and kernel parameter gamma gamma of the DELM model are optimally adjusted by the use of COA. The performance validation of the ICOA-DLDRD method occurs utilizing the benchmark MESSIDOR dataset and the outcomes reported the betterment of the ICOA-DLDRD approach on the recent methods with maximum accuracy of 99.65%.

Automated summary

Diabetic retinopathy is a major cause of preventable blindness for diabetic patients. Regular retinal screening is recommended to detect diabetic retinopathy at an early stage. This study presents an intelligent algorithm based on deep learning for the detection and grading of diabetic retinopathy using retinal fundus images.