Malaria parasite detection in Red Blood Cells with rouleaux formation morphology using YOLOv9

Malaria is endemic in poverty-stricken regions of the world, and most diagnosis reveal comorbidity with other infectious diseases some of which manifest as a deformity of the structural arrangement of the Red Blood Cells (RBCs) during thin blood smear microscopy. This common occurring deformity is termed rouleaux formation, and it is the stacking together of RBCs like chains of coins. The presence of rouleaux formation indicates either a bacterial infection, connective tissue disease, chronic liver disease, multiple myeloma or diabetes among others, it is a highly common occurrence in malaria infected patients and according to the international council for standardization of hematology (ICSH), microscopists are mandated to report its presence. Hence to develop unbiased automated malaria diagnostic systems capable of being deployed in malaria endemic regions, these systems need to be capable of identifying rouleaux formation and detecting malaria parasite within such type of RBC. Thus, this study developed a thin blood smear dataset with rouleaux formation RBCs infected with two species of malaria parasite: plasmodium falciparum and plasmodium malariae. YOLOv9s architecture was used to benchmark the dataset for the detection of plasmodium parasites and white blood cells in the developed dataset. Comparing the effect of using pretrained weights, YOLOv9s trained from scratch achieved a Precision, Recall and mAP50 of 75.4 %, 76.6 % and 80.3 % while YOLOv9s pretrained on the MS COCO dataset recorded an improvement in performance metrics with an increase in Precision by 0.4 %, an increase in Recall by 5.4 % and an increase in mAP50 by 2.5 %