Design of a light-weighted model for enhancement of Malaria Parasite Detection

One of the most serious public health issues in the world is malaria. In many underdevelopednations, it is a major source of disease and mortality for children and expectant mothers whichare the most vulnerable populations. Traditional laboratory malaria diagnosis requires a skilledindividual and meticulous examination to distinguish between healthy and infected red bloodcells (RBCs). The traditional method of carrying out this operation involves a lot of manuallabor that must be done by a human and requires a lot of time and resources. Cognitive computingand machine learning techniques have advanced, and they are now widely employed in thehealthcare sector to detect and anticipate early disease symptoms. Healthcare providers can makeinformed decisions for patient diagnosis and treatment using early prediction results. As of today,researchers implement the most popular image recognition models such as ResNet۵۰, VGG۱۹,and InceptionV۳, ImageNet to detect parasites; However, these models are weighty as a matterof pre-trained weights and a considerable number of layers. This study explores the use of deeplearning algorithms to achieve not only an efficient model but also a light-weighted architectureby customizing the network with an optimal number of layers and parameters. As a result, theproposed model achieves better accuracy with respect to other research works like ResNet۵۰ andVGG۱۹ while having the fewest network parameters.

کلیدواژه ها:

Malaria parasite detection ، Feature extraction ، Deep learning ، Convolutional NeuralNetworks