Analysis of Convolutional Neural Networks and Shape Features for Detection and Identification of Malaria Parasites on Thin Blood Smears

Abstract

The gold standard for malaria diagnosis still remains to be microscopy. However, cases from remote areas needing immediate diagnosis and treatment can benefit from a faster diagnostic process. Several intelligent systems for malaria diagnosis have been proposed using different computer vision techniques. In this research, models using convolutional neural networks, and a model using extracted shape features are implemented and compared. The CNN models, one trained from scratch and the other utilizing transfer learning, with accuracies of 92.4% and 93.60%, both outperform the shape feature model in malaria parasite recognition.