Optical AI pipeline for automated malaria parasite detection in thin blood smears from Surkhandarya Region
Abstract
Balancing diagnostic accuracy and workload is a critical challenge for microscopy-based malaria diagnosis in peripheral laboratories. This study describes an optical artificial intelligence pipeline for automated detection of malaria parasites in thin blood smears collected in the Surkhandarya Region of Uzbekistan. Digitised bright-field images of Giemsa-stained thin smears were acquired with a low-cost slide scanner attached to routine laboratory microscopes and uploaded to a central analysis server. The pipeline integrates colour normalisation, artefact suppression, single-cell segmentation and a convolutional neural network classifier trained to distinguish infected and non-infected erythrocytes and to assign parasite life stages. Diagnostic performance was evaluated against expert microscopists and rapid diagnostic tests, using sensitivity, specificity and slide-level parasite density estimates as endpoints. The artificial intelligence system achieved high accuracy while substantially reducing manual screening time, supporting scalable, quality-assured microscopy for malaria surveillance and case management in a resource-constrained Central Asian setting across diverse local transmission settings.