CellScanner: A High-precision Migratory Cell Recognition Model Based On YOLO And RESNE

Pyroptotic cells are a form of cell death,and the release of pro-inflammatory factors during pyroptosis can recruit immune cells to kill inflamed tissue.Studying the proportion of pyroptotic cells identified by clinical medical imaging can help diagnose and prevent various inflammatory diseases,which has important scientific research significance and clinical application value.The traditional method of identifying pyroptotic cells one by one is prone to misjudgment and is time-consuming and laborintensive.Therefore,it is urgent to develop a tool for automatically identifying cell morphology to reduce human error and workload of manual identification.At the same time,clinical workers often lack the time and proficiency to deploy a deep learning environment.The current deep learning models for medical imaging often only focus on establishing the model,without considering the actual needs and pain points of clinical workers.Therefore,this paper aims to establish an accurate tool.In addition to the pyroptotic cell recognition model,a simple and convenient pyroptotic cell recognition platform was built,so that clinicians can use this model for clinical research and analysis without incurring learning costs.In this paper,the main research objective is to identify pyroptotic cells.Cell Scanner is developed by integrating YOLOv5 and RESNET,and combined with YOLOv5 backbone and RESNET Classifier,it selects and identifies pyroptotic cells in the image with high precision.To address clinical tasks similar to pyroptotic cell identification,this paper establishes the Cell Scanner online platform for online identification of pyroptotic cells and fine-tuning.Its main functions include the prediction function for pyroptotic cells,the function of pre-segmenting cells with HE staining images,and the Cell fine-tuning model training function.Cell Scanner solves problems such as batch effects in shooting and training target differences,which brings great convenience to clinical tasks.Through testing,the fine-tuned online model for centroblasts in lymphoma has reached the expected level,with an average accuracy rate of 92.1% under the condition of Io U threshold of 0.5,and an average accuracy rate of54.1% for Io U threshold from 0.05 to 0.95.The Cell Scanner model can accurately identify specific cells and distinguish differences between cells,and its online platform function can provide great convenience for clinical research.