COMPARISON OF DIFFERENT YOLO ARCHITECTURES IN MICROSCOPIC CELL COUNTING USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

As automation continues to advance rapidly, cell detection and counting plays a significant, crucial role in medical image processing’s automatic analysis. Accurately detecting and locating cells seems to be an extremely difficult task given their size, structure, and adherence to one another. In order to avoid time-consuming manual work, researchers are ever so eager in trying to implement and experiment with new technologies, to achieve a highly performative automatic cell segmentation. Nevertheless, due to different image acquisition techniques, non-uniform backgrounds, variations in cell shapes or size, morphological properties and many other factors, obtaining perfect results is not always so easy. In this paper we study a convolutional neural network approach using YOLOv5 and YOLOv6 architectures for cell segmentation in microscopic images.