Modified YOLOv5 for Blood Cell Counting
Published in RIVF, 2022
The hematology of the human is important information in diagnosing health related problems and it is widely used in medical field. From the complete blood count (CBC) test, the number of blood cell can be determined, and the shape of them can also be used to predict health complications. Besides the conventional methods, deep learning is being applied to count the number of blood cells including red, white blood cell and platelet. In this study, we proposed a method to count blood cell using modified YOLOv5 model, specially by adding Convolution Block Attention Module (CBAM). Experimental results shown that the proposed model CBAM-YOLOv5 achieved mAP@0.5 at 0.878 with confidence score of 0.001 and IoU=0.6 when counting all blood cells. In addition, the proposed model using MobileNetv3 as a backbone achieve accuracy 99.0%, 99.7 % and 99.0% in counting red blood cell, white blood cell and platelet, respectively.