New Normal and Abnormal Red Blood Cells Features for Improved Classification

This paper focused obtaining new features for improved classification of red blood cells (RBCs). RBCs varies according to shapes, colors and sizes. Abnormal RBCs may be caused by anemia. Abnormal RBCs has great similarities among each other causing difficulties in medical diagnosis. In this work, spatial, spectral statistical features and geometrical features of RBCs are extracted from 1000 normal and abnormal RBCs. The extracted features are reduced using Principal Component Analysis (PCA) and tested with different types of machine learning algorithms for classification. Classifications were evaluated for high sensitivity, specificity, and kappa statistical parameters. The classifications yielded accuracy rates of 97.9%, 98% and 98% for discriminative (SVM), generative (RBFNN) and clustering (K-NN) algorithm respectively, which is an improvement over previous works