Ocular disease detection from multiple informatics domains

Computer aided detection for automatic ocular disease detection is an important area of research. As different ocular diseases possess different characteristics and present at different locations within the eye, it is difficult to find a common way to effectively handle each ocular disease. To solve this problem, we propose a unified Multiple Kernel Learning framework called MKL^clm to detect ocular diseases, based on the existence of multiple informatics domains. Our framework is capable to learn a robust predictive model by effectively integrating discriminative knowledge from different informatics domains and incorporating pre-learned Support Vector Machine (SVM) classifiers simultaneously. We validate MKL^clm by conducting extensive experiments for three leading ocular diseases: glaucoma, age-related macular degeneration and pathological myopia. Experimental results show that MKL^clm is significantly better than the standard SVMs using data from individual domains and the traditional MKL method.