Embedded deep-learning based sample-to-answer device for on-site malaria diagnosis

Abstract

AbstractIn response to the ongoing global health problem caused by malaria, especially in resource-limited settings, digital microscopy must be improved to overcome the limitations associated with manual microscopy. In order to present a malaria diagnosis method that is not only accurate at the cell level but also clinically performs well, improvements in deep-learning algorithms and consistent staining results are necessary. The device employs a solid hydrogel staining method for consistent blood film preparation, eliminating the need for complex equipment and liquid reagent maintenance. By leveraging deformable staining patches, the miLab™ ensures consistent, high-quality reproducible blood films can be made across various hematocrits. Embedded deep-learning enables the miLab™ to detect and classify malaria parasites from the autofocused images of stained blood cells by internal optical system, achieving a high correlation with manual microscopy images. This innovation not only minimizes human error but also facilitates remote assistance and review by experts through digital image transmission, revolutionizing the landscape of on-site malaria diagnosis. The miLab™ algorithm for malaria detection shows a total accuracy of 98.83% for infected RBC classification. Clinical validation in Malawi demonstrates an overall percent agreement of 92.21%, highlighting the miLab™’s potential as a reliable and efficient tool for decentralized malaria diagnosis.