Sickle cell anemia detection using convolutional neural network

Sickle cell anemia (SCA) is a genetic condition caused by a mutation in one of the genes that produces hemoglobin. This causes the otherwise circular RBCs to become sickle shaped and obstruct the blood vessels. Sickle Cell disease (SCD) deteriorates the oxygen supply to different parts of the body and leads to severe anemia. Although SCD is not curable, patient's life expectancy and quality can be improved if the disease is diagnosed early and a proper medication and care is made available. Hence precise detection and classification of blood cells is important. Manual classification is tedious and requires expertise. Automatic classification using image processing and machine learning is feature dependent, inefficient and sensitive to change in color, shape and size of the blood cells. Hence this research proposes the use of deep neural network to detect the existence of sickle-shaped cells in blood samples so that a precise diagnosis of the disease can be done. Here RBCs are classified in one of the three classes: Normal, sickle shaped and other blood components. Histogram equalization is used for preprocessing and data augmentation technique is used to tackle the shortage of dataset. We achieved a testing accuracy of 94.57% with a CNN consisting of only 5 convolution layers.