Automatic Detection and Counting of Malaria Parasite-Infected Blood Cells

Abstract

In this paper, we present a technique for automatic detection and counting of Plasmodium vivax-infected red blood cells by means of a convolutional neural network and a feature- based counting process. Current approaches for object detection or counting often rely on prior knowledge of certain salient features of the to-be-identified objects or require time-consuming pre-processing. For this reason, many detection problems, for example infected cell counting, remain a manual task for trained professionals, leading to potentially high amounts of time between infection and diagnosis and treatment, which, in turn, can have lethal consequences. Using the BBBC041 data set, we annotated the ground truth (GT) of infected cells with circles in each image and then trained a convolutional neural network to predict these GTs from previously unseen cell images. Subsequently, the algorithm computes the number of cells using Canny edge detection and circular Hough Transform. © 2020, Springer Nature Singapore Pte Ltd.