Deep Convolutional Neural Network Model for Detection of Sickle Cell Anemia in Peripheral Blood Images

Abstract

Emmanuel Gbenga Dada*, David Opeoluwa Oyewola and Stephen Bassi Joseph

Sickle Cell Disease (SCD) is a disorder of red blood cells (RBC). The number of SCD patients is rising daily. The lifespan of people is reduced by this deadly disease. Statistics show that over twenty five percent of people living in the Central and West Africa region are suffering from this malady. Many of the nations in this part of the world are deficient in the essential means of detecting and treating several illnesses of which SCD is one of them. Infant mortality rates are considerably greater in these countries. The conventional techniques for SCD diagnosis are expensive, error-prone, time consuming, and require the services of medical experts. As a result, there is a pressing need to develop cost-effective and controllable approaches for the early detection and diagnosis of SCD. This paper presents novel techniques that use Plain Convolution Neural Networks (PCNN) with 15 layers and 48 layers, data augmentation of Plain Convolution Network with 48 layers (DAPN-48), Very Deep Convolutional Networks for Large Scale Image Recognition with 19 layers (VGG19), and Residual Networks with 50 layers (RESNET-50) for detecting SCD from peripheral blood image samples. Results obtained from our experiments indicated that PCNN-15 and DAPN-48 outperform PCNN-48 with sensitivity and balanced Accuracy between 99-100%. A comparison was made between the performance of PCNN-15, PCNN-48, DAPN-48, VGG19 and RESNET-50. The results attained by the proposed approaches demonstrated that our techniques are appropriate for the diagnosis of SCD, and thereby recommended for application to sickle cell image detection.