Deep Learning Based Automatic Identification of Malaria Infected Cells
TL;DR
The findings indicate that DenseNet121 greatly exceeds the performance of ResNet50 with a test accuracy of 92.15% as opposed to the 70.19% accuracy of ResNet50, thereby showing the benefits of densely connected feature re-use in detecting minute intracellular parasitic features.
Abstract
The study focuses on the application of artificial intelligence as a way of classifying malaria-infected red blood cells accurately and transparently. In particular, the classic method of diagnosing malaria is centered on a painstaking examination of the blood smear by qualified personnel which still proves effective, but time- and labor-consuming, as well as subjective, and dependent on the presence of professionals. Nevertheless, deep learning provides an information-driven approach that allows for the identification of unique visual patterns in labeled cells. Thus, the aim of this study is to develop and assess various models based on convolutional neural networks that would help distinguish the categories of red blood cells based on blood smear images into Parasitized and Uninfected ones. The methodology implies collection of the data set with images of red blood cells, their processing, and transfer learning on the basis of models like ResNet and DenseNet. Additionally, the use of the Grad-CAM technology helps make the diagnostic system even clearer. The findings indicate that DenseNet121 greatly exceeds the performance of ResNet50 with a test accuracy of 92.15% as opposed to the 70.19% accuracy of ResNet50, thereby showing the benefits of densely connected feature re-use in detecting minute intracellular parasitic features.