摘要
背景:基于超声图像的卷积神经网络(Convolutional neural network,CNN)的诊断模型是目前临床诊断乳腺癌良恶性的有效手段。但相比良恶性二分类诊断,乳腺癌病理细分多分类诊断可能更利于临床更精准治疗,更具临床价值。目的:基于超声图像构建乳腺癌CNN病理类型诊断模型,并分析其临床诊断价值。方法:纳入我院2016年1月至2022年1月经病理学检查确诊的432例乳腺癌患者,其中262例为原位癌,170例为浸润癌。收集所有患者超声图像特征等资料作为建模数据集,并将建模数据按7:3的比例划分为训练集和验证集。构建基于超声图像的乳腺癌患者病理类型的CNN诊断模型及Logistic回归诊断模型,并使用ROC曲线下面积来评价两种诊断模型的性能。结果:共筛选出形态、毛刺蟹足、Adler血流分级、腋窝淋巴结转移、后方回声衰减5个超声图像作为建模指标,通过分析训练集与测试集数据流分别建立CNN诊断模型及Logistic回归诊断模型,CNN诊断模型的灵敏度、特异度、准确度AUC曲线下面积CNN诊断模型均高于Logistic回归诊断模型。结论:基于超声图像的乳腺癌CNN病理诊断模型在诊断乳腺原位癌及浸润癌的临床诊断价值高于Logistic回归诊断模型,对临床乳腺癌患者病理类型诊断具有一定的辅助价值。
BACKGROUND:The diagnostic model of Convolutional neural network(CNN)based on ultrasound images is an effective means for clinical diagnosis of breast cancer.However,compared with the two-category diagnosis of benign and malignant breast cancer,the pathological subdivision and multi-category diagnosis of breast cancer may be more conducive to more accurate clinical treatment and have more clinical value.Objective:To construct CNN pathological diagnosis model of breast cancer based on ultrasound images and analyze its clinical diagnostic value.Methods:A total of 432 patients with breast cancer diagnosed by pathological examination in our hospital from January 2016 to January 2022 were enrolled,of which 262 were in situ cancer and 170 were invasive cancer.Collect all patient ultrasound image features and other data as a modeling data set,and divide the modeling data into a training set and a validation set in a ratio of 7:3.A neural network diagnostic model of breast cancer patient pathological types based on ultrasound images was constructed,and its performance in training and validation sets was compared.Results:Five ultrasound images,including morphology,hairy crab feet,Adler blood flow classification,axillary lymph node metastasis and posterior echo attenuation,were selected as modeling indexes.The CNN diagnostic model and Logistic regression diagnostic model were established by analyzing the data streams of the training set and the test set,respectively.The sensitivity,specificity and accuracy of the CNN diagnostic model were higher than those of the Logistic regression diagnostic model.Conclusion:The CNN pathological diagnosis model of breast cancer based on ultrasound images has higher clinical diagnostic value in diagnosing breast cancer in situ and invasive cancer than Logistic regression diagnosis model,and it has certain auxiliary value in clinical diagnosis of pathological types of breast cancer patients.
作者
於子扬
唐维
杨丽
YU Ziyang;TANG Wei;YANG Li
出处
《生命科学仪器》
2023年第1期63-69,共7页
Life Science Instruments
关键词
卷积神经网络模型
超声图像
乳腺癌
病理类型
neural network model
ultrasound image
breast cancer
pathological type
