Prediction of malaria plasmodium stage and type through object detection

Abstract

The use of image processing, artificial intelligence in general, and objects detection in particular for the diagnosis of malaria is increasingly remarkable. In the present work, we suggested a comparison of two objects detection models which is not only capable of detecting the infected blood smears cells but also it enables to distinguish the plasmodium different species and the parasitic stage of malaria. To create our two models, transfer learning from the Faster-RCNN and YOLO models have been used with the MP-IDB database (Malaria Parasite Image Database for Image Processing and Analysis). Then, we have followed three main steps to develop our approach of objects detection for malaria: the creation of annotated databases, image preprocessing, and fine-tuning the two pretrained models on our annotated database. The Faster R-CNNbased model produced better results than the Yolo-based one, with a mAP @.50IOU of 0.76 versus 0.3.