Automated Malaria Detection and Severity Prediction using Deep Learning: A VGG19-based Approach

This is a cause for life threatening disease of Plasmodium parasites, Malaria. Effective treatment and management require early detection and severity prediction. The main contribution of this paper is to present a deep learning based approach for automated detection and prediction of malaria severity from blood smear images. To classify images between infected and uninfected red blood cells, we use VGG19, a pre trained Convolutional Neural Network (CNN). Data augmentation techniques like rotation, zoom, brightness adjustments are used to augment the dataset and improve the model generalization in the proposed model. Moreover, the severity level of malaria is predicted by percentage of infected cell and is classified as mild, moderate or severe. To overcome the limitations of preceding methods in the context of mainly detecting without predicting severity, we propose an approach. This model is evaluated on a well annotated Kaggle dataset and performs well. The developed methodology may offer an efficient and scalable option for malaria diagnosis and assessment of severity that may be useful in resource limited settings.