基于深度卷积神经网络的外周血液细胞及抗体检测方法  

Peripheral Blood Cells and Antibodies Detection Approach Based on Deep Convolutional Neural Network

作　　者：李艾彤 李铭诗[2] 李靖鹏 李晨[2] LI Aitong;LI Mingshi;LI Jingpeng;LI Chen(Department of Radiology,The Fourth Affiliated Hospital of China Medical University,Shenyang Liaoning 110032,China;College of Medicine and Biological information Engineering,Northeastern University,Shenyang Liaoning 110167,China)

机构地区：[1]中国医科大学附属第四医院放射诊断科,辽宁沈阳110032 [2]东北大学医学与生物信息工程学院,辽宁沈阳110167

出　　处：《中国医疗设备》2024年第12期22-27,45,共7页China Medical Devices

基　　金：国家自然科学基金重点项目(82220108007)。

摘　　要：目的为了解决目前常用血细胞检测仪器检测耗时长、检测细胞种类少及人工镜检方式耗时耗力的问题。方法为了更快捷地对血细胞进行检测,提供准确的检测报告,减少医生工作量,本研究结合区域卷积神经网络、YOLO、单次多框检测器(Single Shot MultiBox Detector,SSD)等深度学习方法进行血细胞检测,并在实验中选择外周血细胞数据集,应用SSD模型和YOLO系列中的YOLOv5、YOLOX、YOLOv6、YOLOv7共5种网络模型进行训练,对比评估指标并讨论各网络模型的优劣势。结果本研究搭建了精准度更高、运行速度更快的新模型,精准度可达99.3%,单张检测时间为10.3 ms,并且所占内存仅为71.2 MB,超过其他网络模型;本文的消融实验验证了新添加的全连接层模块与泛化模块的实用性。结论该模型能够出色且准确地检测血细胞,且检测速度快、准确率高,同时模型较小,方便使用和维护。Objective To solve the problems of time-consuming detection,few cell types and time-consuming manual microscopic examination methods.Methods In order to detect blood cells more quickly,reduce the workload of doctors and provide accurate reports,this study combined the regional convolutional neural network,YOLO,single shot multiBox detector(SSD)and other deep learning methods to detect blood cells.In this experiment,the peripheral blood cell data set was selected,and the SSD model and five network models in the YOLO series,YOLOv5,YOLOX,YOLOv6 and YOLOv7 were used for training.The advantages and disadvantages of the network were discussed by comparing the evaluation indicators.Results This study built a new model with higher accuracy and faster running speed.The accuracy reached 99.3%,the single-image detection time was 10.3 ms,and the memory occupied was only 71.2 MB,surpassing other network models.An ablation experiment was designed to verify the practicality of the newly added fully connected layer module and generalization module.Conclusion This model can detect blood cells excellently and accurately,with fast detection speed and high accuracy.The model is small and easy to use and maintain.

关 键 词：血细胞检测 目标检测 外周血细胞 深度学习网络模型 消融实验 YOLO 单次多框检测器(SSD) 

分 类 号：R197.39[医药卫生—卫生事业管理] TP391.4[医药卫生—公共卫生与预防医学]