Semi-Supervised Learning Based on Cataract Classification and Grading

Cataract is one of the major causes of blindness and vision impairment. In previous research work which includes fundus image pre-processing, feature extraction and using labeled samples to build classifier to reach the goal that cataract classification and grading automatically. To learn a well performed hypothesis, a large amount of labeled examples are required. The labeled examples are fairly expensive to obtain. In this paper, we utilize semi-supervised learning to build a classifier for automatic classification and grading of cataract, which can reduce the heavy burden on ophthalmologists. Using a large scale of unlabeled examples together with a small part of labeled examples to learn hypothesis is known as semi supervised learning. Many semi-supervised learning algorithms existed at present. We used tri-training which generates three classifiers from the original labeled examples. Then utilizing unlabeled examples to refine initial classifier in an iterate method. Experiments on real word data sets included 476 labeled examples and 4902 unlabeled examples. The empirical experiments are conducted for cataract detection and cataract grading. The best performance of the semi-supervised learning is 100% and 88%. Experiment results provide a bright future in later practical application of classification system of cataract detection and grading. It also illustrates the effectiveness of the proposed approach.