AUTOMATIC CLASSIFICATION OF NORMAL AND ABNORMAL HUMAN RED BLOOD CELL USING GEOMETRIC FEATURES AND FACTORS

Abstract

Computer based diagnosis of approach enables to detect abnormalities of red blood cells. Automated computer-based diagnosis is believed to be fast and accurate if the system is properly developed and verified. The system my help to make early detection of diseases such as malaria and anemia so that suitable follow up and treatment can be done. In this regard, this thesis presents a method for automatic segmentation, features extraction and classification of red blood cells as normal and abnormal using image processing techniques and learning algorithm. To this work, image processing techniques such as binarization, contrast enhancement, noise elimination, morphological operations (fill hole, clear boarder and remove small object), labeling and extraction of features of interest (area, perimeter, major axis length and minor axis length) were done. The red blood cells were mainly classified using discriminating factors (form factor, circularity factor, and deviation factor).The extracted features and factors were used as inputs for the neural network, which classified the RBC images as normal or abnormal. Classification was carried out based on the back propagation learning algorithm, which involved the training of the network using108 normal and 365 abnormal cells from RBC image samples. The classification results categorize red blood cell as normal and abnormal. The results showed that using the proposed ANN classifier, have sensitivity and specificity of 98.9% and 99.1%, respectively. The accuracy (98.9%) of the result is determined by doing comparison with the ground truth data.