Multiclass multilabel ophthalmological fundus image classification based on optimised deep feature space evolutionary model

Primary care doctors have been fighting against ocular illnesses for more than 37% of the world's population. This demonstrates the need for an autonomous and intelligent technological solution to improve the accessibility and convenience of categorising retinal pathology. The lab technician examines around 80 control individuals each day in addition to hospitalised patients for an average of 12 minutes to identify each disease. The study suggests using DFex-BeeHive, or deep feature extraction through the Bee Hive network, to categorise DR lessons across several labels. The evaluation of cutting-edge deep learning, machine learning, and algorithmic techniques is performed using the proposed DFex-BeeHivearchitectureIn order to reduce the inherent multicollinearity in deep learning, the research recommends using CGAN to flatten the distribution function of probabilities and PSO to synchronise the selection of heuristic-based features. The work uses a hybrid approach of CGAN, PSO, and DFex-BeeHive architecture to obtain 98.79% accuracy, 95.99% sensitivity, and 99.79% specificity in the RFMiD dataset and 97.16% accuracy and 96.81% F1 score in the ODIR dataset. In addition to improving classification precision over earlier lesion classifiers, the work reduces computation by 47% when compared to other cutting-edge designs using feature selection techniques.