Detection of Acute Lymphoblastic Leukemia Subtypes using YOLO and Mask R-CNN

White Blood Cell (WBC) detection and counting on Microscopic Blood Cell images automatically can support expertise in diagnosing Acute Lymphoblastic Leukemia (ALL) more easily and quickly. Previous research commonly uses conventional approaches for ALL subtype detection that needs several stages, such as WBC segmentation, touch cell separation, feature extraction, and classification. We present object detection and instance segmentation techniques that require only one learning framework without needing separate stages. In this paper, we compare the performance of the YOLO and Mask R-CNN models for the detection of ALL subtypes using evaluation metrics such as precision, recall, F1, and mAP. The results show that the YOLOv4 outperforms YOLOv5 and Mask R-CNN in the detection of ALL subtypes. The YOLOv4 model has slightly better performance than the YOLOv5 model and Mask R-CNN, with an F1 value of 89.5% and a mAP value of 93.2%, respectively.