DEEP LEARNING FOR OPTICAL MALARIA DIAGNOSIS: A SYSTEMATIC REVIEW OF ALGORITHMIC ADVANCES, DATASET BIASES, AND THE INDIGENOUS DATA GAP

ABSTRACT

Malaria remains a leading cause of mortality in sub-Saharan Africa. While manual microscopy of Giemsa-stained blood films is the diagnostic gold standard, it is labor-intensive and prone to variability.Deep learning, particularly convolutional neural networks (CNNs), has shown strong potential for automated malaria diagnosis; however, clinical translation in African settings remains limited. This systematic review, following PRISMA guidelines, examined 35 studies (2015–2024) evaluating CNN architectures and their training datasets. Findings reveal a critical "indigenous data gap": most models are trained on non-African datasets under controlled conditions, suffering from dataset shift when deployed in low-resource environments. The review also highlights absent gender-disaggregated reporting and minimal engagement with responsible AI principles. We conclude that developing FAIR-compliant indigenous datasets is essential for equitable AI-driven malaria diagnosis in Africa.

Keywords: Deep Learning, Malaria Microscopy, Dataset Shift, Algorithmic Bias, Indigenous Data, CNN