Computer aided diagnosis of nuclear cataract

An approach to automatically diagnose nuclear cataract based on the slit-lamp image is proposed in this paper. Model-based approach is investigated to detect robust lens structure. Based on the detected lens structure, the mean intensity, the color information on the posterior subcapsular reflex and visual axis profile are extracted as the grading features. Support Vector Machine (SVM) regression model is further trained to predict the grades of nuclear cataract automatically. The proposed approach was tested using 900 images and the mean grading error is 0.36. The encouraging results show that it is promising to apply the proposed approach to clinical diagnosis.