Android-Based Malaria Detection Using Deep Learning

The objective of this research is to explore the feasibility of using smartphones for the process of identifying malaria parasites in images of thick blood smears through automation. We have developed a deep learning method that can detect malaria parasites in thick blood smear images and can run on android-based smartphones. The proposed method involves two steps: first, an intensity-based screening process that quickly identifies potential parasite candidates and second, a customized convolutional neural network that classifies each candidate as either infected or uninfected. We have compiled various publicly available datasets and formed a dataset containing 3024 thick smear images for our research. We have trained our proposed deep learning method on this dataset and experimental results show that it is highly effective in discriminating between positive and negative images. The proposed method is achieving high accuracy of 94.19%, area under the curve (AUC) is 94.13% with 91.22% sensitivity, 97.04% specificity, 93.78% precision, and 98.11% of negative predictive value.