Fundus Image Based Cataract Classification

Cataract is one of the leading causes of visual impairment worldwide. People with cataracts often suffer a lot in many aspects of daily life. Although early treatment can reduce the sufferings of cataract patients and prevent visual impairment turning to blindness, people in less developed areas still can't get timely treatment because of poor eye care services or lack of professional ophthalmologists. Besides, the present commonly used methods for cataract diagnosis, clinical assessment and photographic grading, need to be operated at a slit lamp by ophthalmologists, which are complicated and expensive for many patients. So reducing the cost and simplifying the process of early cataract diagnosis is of great importance. In this paper, we proposed a fundus image based cataract classification method by using pattern recognition, which can be used in early screening of cataract. By calculating the 2-dimensional discrete Fourier transform of a fundus image and using the calculated spectrum as features, a cataract classification and grading method is carried out by using the linear discriminant analysis promoted with the AdaBoost algorithm as the classifier. A preliminary test is implemented on an image sample set including 460 fundus images that normal, mild, moderate and severe cataract images are 158, 137, 86 and 79 respectively. Correspondingly, the two-class and four-class classification accuracy for our proposed method are 95.22% and 81.52%. We believe that our proposed method has a great potential in practical applications.