Detection of Anemia by Digital Photography of Palpebral Conjunctiva

Abstract

Medical image processing provides state of the art techniques for the automatic diagnostic system to different diseases and is very common now days. These types of systems provide cost effective health care facilities to patients. Anemia globally is prevalent most in developing countries. Globally, anemia is affecting 1.62 billion people, which comprises of 24.8% of the population. The maximum occurrence of anemia is in preschool age children and the lowest occurrence of anemia is in men. But, the greatest number of individuals, with the population group affected is non-pregnant women. Like other developing countries, Pakistan has been facing problem of Anemia which is very common among women and children. Iron deficiency is the prevalent most nutritional deficiency amongst many newborns and children consisting of 83% of all anemia. In this thesis a non-invasive automatic detection of anemia is proposed by using image processing and pattern recognition techniques. An image of face with visible conjunctiva is given as input to the system and eye region is detected using Viola and Jones algorithm. The method then enhances the contrast of image in preprocessing step. The conjunctiva segmentation is performed by using K-means clustering. Color based feature vector is then computed and by using these feature set, SVM classifier is applied in two stages to classify between anemic and non-anemic images and then further into mild and severe anemic images. Two camps have been arranged locally to gather patients’ data containing the images and clinical HB values. The results of proposed system are correlated with the clinical values and the results shows that this non-invasive method can be used as an initial guess for anemia detection.