基于深度学习的肺癌病理图像分类器设计
Design of pathological image classifier for lung cancer based on deep learning
朱煜尔 1刘晓帆2
作者信息
- 1. 新乡医学院医学工程学院,新乡 453000
- 2. 新乡医学院三全学院智能医学工程学院,新乡 453000
- 折叠
摘要
病理学检查是医生确定肿瘤是否发生癌变的"金标准",但由于肺癌病理组织具有多种亚型,医生需要反复大量阅片才能最终给出医学诊断,不仅耗时且易出错.因此,本文借助深度学习进行肺癌病理组织亚型分类研究.通过对数据库数据进行预处理,找出特征值,使用不同层深的ResNet算法构建肺癌病理图像类别分类器,模型参数调整到最优后,对比训练ResNet18、ResNet34、ResNet50 三个不同层次网络模型,分析模型的accuracy、recall、F1-score和算术平均值等评价指标,其中ResNet34 模型指标最佳,对肺癌病理图像的分类效果最好.
Abstract
Pathological examination is the"gold standard"for doctors to determine whether a tumor has undergone cancerous transformation.However,due to the multiple subtypes of lung cancer pathological tissue,doctors need to repeatedly review a large number of films in order to finally give a medical diagnosis,which is not only time-consuming but also prone to errors.Therefore,this article utilizes deep learning to study the subtype classification of lung cancer pathological tissues.By preprocessing the database data,finding the feature values,and using different layers of ResNet algorithms to construct lung cancer pathology image classification models,after adjusting the model parameters to the optimal value,we compared the training results of three different network models:ResNet18,ResNet34,and ResNet50.We analyzed the accuracy,recall,F1-score,and arithmetic mean of the models,and found that the ResNet34 model had the best performance in terms of classification accuracy for lung cancer pathology images.
关键词
肺癌亚型/深度学习/ResNet模型/病理图像/Adamax优化器Key words
Lung cancer subtypes/Deep learning/ResNet model/Pathological images/Adamax optimizer引用本文复制引用
出版年
2024
NETL
NSTL
万方数据