Nuclear Cataract Database for Biomedical and Machine Learning Applications

A cataract is a medical condition causing an opacity in the ocular nucleus due to various factors such as age and diseases. Starting from traditional image processing techniques for processing and extracting relevant features, using computational intelligence methods is essential to help experts in the medical pre-diagnosis for automatic classification and grading of the disease. However, the learning capabilities of such automated processes rely considerably upon the availability of adequately-labeled databases approved by medical experts. Considering the shortage of available public databases for implementing potential algorithms such as Deep Learning, this work presents a new nuclear cataract database composed of 1437 labeled images for multi-level classification according to the LOCS III system. The images were obtained and correctly classified by experts from an ophthalmologic medical center in Mexico City. Also, our research compares relevant Machine Learning algorithms employed for medical images like Support Vector Machines, Deep Learning structures like GoogLeNet, and our proposed Deep Learning Structure with the highest classification rates for the two and multiple cataract levels according to LOCS III.