Malarial Parasite Detection Based on Smartphone Microscopic Imaging Using Deep Learning Approach

Abstract

Malaria is a dangerous and life-risking disease and sometimes leads to death. Microscopy examination was used for diagnosing malaria-infected cells in the early days. Due to a large number of samples for analysis and complexity of time, it may lead to false detection. More time consumption and false detection resulted in a great need for automated parasite detection systems. The proposed work aims to detect the malaria-infected images from microscopic blood smear images which are acquired by smartphones. Detection of malaria-infected images is done by using a convolutional neural network model called ResNet. In the proposed work, the deep learning approach is used to provide a more reliable diagnosis, specifically in resource-limited areas and it also reduces the cost of diagnosis. As the microscopic blood smear images are acquired by smartphones, it provides cost-effectiveness and is easy for gathering datasets with less time. It can also quickly transfer the blood smear images for early diagnosis. In the proposed work, the images are passed through a convolutional layer consisting of residual units which are defined by ReLu and Batch normalization. Finally, it is proceeded by fully connected layer to give the predicted output either malarial infected or uninfected images. The training and validation accuracy and loss graphs have been plotted and the performance metrics of the model have been evaluated.