Automatic parasitemia level quantification at blood slide images for malaria drug discovery

Thick blood smear (TBS) remains the gold standard for evaluating new antimalarial effects against field isolates in schizont maturation assays, one of the most common in vitro techniques employed for drug discovery. Despite its low-cost and wide availability, TBS is a laborious, time-consuming process that heavily relies on experienced parasitologists, trained personnel, and facilities, limiting its use in resource-limited environments. Therefore, we developed a deep-learning-based (DL) autonomous system to quantify parasitemia levels at blood slide images (BSI), a fast and low-cost automatic solution that will potentially improve parasitologists' efficiency and overcome TBS drawbacks. To do so, we have collected 2,000 blood slide images from patients infected with P. falciparum and P. vivax. Then, all parasites were bounding-box labeled using the tool SuperAnnotate. Finally, DL models were trained to detect and classify parasites by stage using YOLO v3. Our models achieved a high mean average precision (mAP) of 0.86.