Abstract:
Malarial Retinopathy is an eye disease which affects retina of a patient suffering from malaria. The changes in vision are only observable at advanced stages of malaria. The major abnormalities seen in malarial retinopathy are retinal whitening, cotton wool spots and retinal hemorrhages. Retinal whitening is a unique sign of malarial retinopathy and further classification into macular whitening or peripheral whitening, reduces the chance of vision loss and cause of cerebral malaria. The proposed system detects the abnormal regions by using some image processing techniques with a combination of machine learning tools for feature set formation and selection. The designed algorithm detects white patches in the image by extracting no of features at image level and selects top features for classification whereas for hemorrhages and cotton wool spots pixel based method is applied to segment the candidate regions and after post-processing and thresholding blot shaped hemorrhages and cotton wool spots are detected. The evaluation of implemented system has been done by on locally gathered data set of malarial patients from AFIO, hospital Rawalpindi. The results are computed on basis of different performance measures and achieved an average accuracy of 90% and 97.3% for retinal whitening and retinal hemorrhages respectively and cotton wool spots are accurately detected with 82.21% sensitivity. The suggested automated system will help in early detection of malarial retinopathy to prevent from vision loss and other severe disease.
