CATARACT DETECTION AND GRADING BASED ON COMBINATION OF DEEP CONVOLUTIONAL NEURAL NETWORK AND RANDOM FORESTS

Cataract is one of the most common eye diseases which leads to visual impairment and is the main cause of blindness. Early intervention and timely treatment can largely avoid cataract blindness. Cataract grading based on fundus images by artificial intelligence algorithms is a feasible method to assistant doctors to diagnose cataracts more effectively. In this paper, a method that Deep Convolutional Neural Network (DCNN) combined with Random Forests (RE) is proposed for six-level cataract grading. In this method, DCNN consists of three modules for feature extraction at different levels on fundus images, while RE implements more elaborate six-level cataract grading based on the feature datasets generated by DCNN. The six-level grading allows doctors to understand the patient's condition more accurately than four-level grading. The accuracy of six-level grading achieved by the proposed method is up to 90.69% on average, with superiority in specificity and sensitivity indicators. Our experimental results also show that RF improves the grading accuracy and reduces the concussion of DCNN on small datasets.