Comparative Evaluation of YOLOv5, YOLOv8, and YOLOv11 for Real-Time Leukemia Cell Detection in Blood Smear Images

Abstract - This study compares and implements three deep learning-based object detection models, YOLOv5, YOLOv8, and YOLOv11, for the purpose of detecting leukemia cells in blood smear samples using image analysis via light microscopy. The pri- mary aim of the study was to investigate the performance of the three models to assess their worth in targeting abnormal leukemia cells in peripheral blood, with the metrics of precision, recall, mean Average Precision (mAP) and inference time considered using the ALL-IDB1 dataset. YOLOv5 was selected to deploy the model for real-time use, primarily because it is lightweight, and has a high recall—theoretically at 1.0 recall, which is ideal in order to limit false negatives in medical diagnosis; despite YOLOv11 yielded higher accuracy and recall. A Flask-based web application was developed for ease of use that utilized supported standard upload of images, displayed the prediction on the input image, and monitored real-time detection while integrating YOLOv5 for the detection of leukemia cells in microscopic blood smear images. Through careful testing and system integration, a conclusion was outlined that considered the potential of YOLO- based models in improving early and accurate diagnosis of leukemia, while providing a novel and exciting platform for diagnosis/image analysis via real time methods. Key Words: Leukemia Detection, YOLO, Object Detection, Deep Learning, Flask Application, Blood Smear Analysis, Medical Imaging