Interpretable AI Ensemble Classifies Malaria Species in Ethiopian Blood-Smear Images
Published in Artificial Intelligence, Journals, Tools.
Published October 5, 2026. Study publication date: September 26, 2026.
A peer-reviewed BMC Medical Imaging study describes an AI ensemble for distinguishing malaria parasite species in thin blood-smear images collected in Ethiopia. The accepted article is available online ahead of its final version.
The researchers combined a custom convolutional neural network with five established image-classification architectures using soft voting. Development and evaluation used 37,200 Giemsa-stained images. Reported classification accuracy ranged from 99.87% to 99.987%, with a macro-averaged F1 score of 0.9822.
Grad-CAM heatmaps were used to inspect which image regions influenced predictions. These high scores reflect performance on the study dataset and do not establish equivalent performance in independent clinical deployment. The authors identify reliance on thin smears and absence of validation for mixed-species infections as limitations; mobile deployment remains future work.