Malaria parasites detection and identification using object detectors based on deep neural networks: a wide comparative analysis

Malaria is an infectious disease transmitted by the bite of the female Anopheles mosquito, infected by Plasmodium spp. Early diagnosis and prompt and effective treatment are needed to avoid anaemia, organ failure, and death. Manual microscopy is the primary technique used for diagnosing malaria. However, this exam is labour-intensive and requires qualified personnel. In order to alleviate these difficulties, researchers are using computer vision concepts to detect and classify cells infected by the Plasmodium spp. This paper presents a comparative study of seven object detectors based on deep neural networks. We evaluated four versions of the You Only Look Once (YOLO), Single Shot MultiBox Detector, EfficientDet, and Faster R-CNN convolutional neural networks to detect and classify the malaria parasite in six different scenarios. We used four public image datasets, and the YOLOv5 model outperformed the other evaluated models. Their results had similar performance to the state-of-the-art works demonstrating the feasibility and effectiveness of YOLOv5 to detect and classify malaria parasites in thin blood smear images with high precision and sensitivity.