Pathogen Of Malaria Detection from Thick Smears of Blood Using Dl Algorithms Over Smart Mobile Devices

Abstract

The focus of this investigation lies in the realm of automated malaria parasite detection within thick blood smears, with a particular emphasis on the integration of smartphones as the primary platform for its deployment. In our pioneering endeavor, we introduce a novel deep learning methodology, a groundbreaking innovation engineered for the purpose of malaria parasite detection in thick blood smear images, meticulously designed to operate seamlessly on smartphone devices. The methodology that underpins our approach consists of a two-tiered processing paradigm. Initially, we embrace the utilization of an intensity-based Iterative Global Minimum Screening (IGMS), a rapid and meticulous mechanism that screens thick smear images, sifting through the visual landscape to identify prospective parasite candidates. Following this initial screening, we enlist the expertise of a customized Convolutional Neural Network (CNN), purpose-built for the classification of each candidate, deftly categorizing them as either parasites or belonging to the background. As a hallmark of our commitment to the advancement of scientific knowledge and research, we accompany this paper with the gracious offering of a dataset, containing 1819 thick smear images culled from the lives of 150 patients. This repository, made freely accessible to the global research community, stands as a testament to our dedication to fostering collaboration and innovation in the realm of malaria parasite detection. Our rigorous scientific inquiry leverages this dataset for both the training and testing of our deep learning methodology, as meticulously documented within the pages of this paper. The results of our comprehensive endeavor are artfully summarized through a patient-level five-fold cross-evaluation. This evaluation bears witness to the efficacy of our customized CNN model, as it adeptly discriminates between positive (parasitic) and negative image patches. The litmus test is exemplified through a suite of performance indicators, each revealing a facet of the model's prowess in capturing the nuances of the data landscape.