Improved ResNet-Based Image Classification Technique for Malaria Detection

Malaria is a communicable disease instigated by mosquitoes and is life-threatening. As per the World Health Organization (WHO), there are 216 million cases of malaria spread across 91 countries. According to the report dated December 2019, there are 228 million people affected by malaria in 2018 in contrast to 231 million ones in 2017. The traditional way of diagnosing malaria is by visually inspecting blood samples of Red Blood Cells (RBCs) infected by the parasite under a microscope. However, this method is time-consuming as the diagnosis is entirely manual and is based on the expertise of the examiner. Other alternate diagnostic methods include Automatic Image Recognition (AIR) technologies based on Machine Learning (ML). These methods when applied to blood smear yield non-optimal results. To overcome the drawbacks associated with these techniques, the proposed system uses a Residual Neural Network (ResNet) architecture for detecting malaria from blood sample images. Classification is performed on the augmented image dataset to achieve more accuracy. The model offers 93.25% diagnostic accuracy on the training data and 87.35% accuracy on the validation data.