基于T2WI和ADC图像放射组学特征的机器学习模型鉴别早期宫颈癌和慢性宫颈炎  

The Value of Machine Learning Models Based on T2WI and ADC Imaging Radiomics Features in Differentiating Early Cervical Cancer from Chronic Cervicitis

作　　者：杨守义 吴卓翰 韩朝钢 侯悦 柳佳丽 马嘉敏 邓义 YANG Shouyi;WU Zhuohan;HAN Chaogang;HOU Yue;LIU Jiali;MA Jiamin;DENG Yi(Department of Radiology,Shaoguan Maternal and Child Health Hospital,Shaoguan,Guangdong 512000,China)

机构地区：[1]韶关市妇幼保健院放射科,广东韶关512000

出　　处：《生物医学工程学进展》2025年第1期25-30,共6页Progress in Biomedical Engineering

摘　　要：目的探讨基于T2WI和ADC图像放射组学特征的机器学习模型在鉴别早期宫颈癌和慢性宫颈炎中的价值。方法回顾性分析2019年9月至2023年2月韶关市妇幼保健院病理确诊的宫颈病变患者,其中早期宫颈癌患者34例(阳性组),慢性宫颈炎患者46例(阴性组)。患者被分为一个训练集(56例,其中24例阳性和32例阴性)和一个独立的测试集(24例,其中10例阳性和14例阴性)。收集每例患者的T2WI和ADC图像,通过3D Slicer 5.4.0软件和PyRadiomics软件模块从宫颈的容积感兴趣区中提取837个放射组学特征。采用3种数据归一化方法、2种数据降维方法、4种特征选择方法和10种机器学习模型分类器。在建模过程中,最优特征参数被设置为1~12。对训练集进行10倍交叉验证,以确定模型的超参数。采用独立测试集的AUC评价模型性能。结果总共建立了2880个机器学习模型,其中Mean_PCC_RFE_5_SVM模型的预测性能最佳。结论基于T2WI和ADC图像放射组学特征的机器学习模型具有区分早期宫颈癌和慢性宫颈炎的应用价值,与其他机器学习模型相比,支持向量机具有更高的诊断效率。Objective To explore the value of machine learning models based on T2WI and ADC imaging radiomics features in differentiating early cervical cancer from chronic cervicitis.Methods A retrospective analysis was conducted on 80 patients with pathologically confirmed cervical lesions in Shaoguan Maternal and Child Health Hospital from September 2019 to February 2023,including 34 with early cervical cancer(positive group)and 46 with chronic cervicitis(negative group).The patients were split into a training set(56 cases:24 positive and 32 negative),and an independent test set(24 cases:10 positive and 14 negative).T2WI and ADC images of each patient were obtained,and a total of 837 radiomics features were extracted from cervical VOIs by 3D slicer software and the PyRadiomics software module.Three data-normalization methods,two dimensionality-reduction methods,four feature selection methods and ten machine learning methods were used.The optimal number of features in the modeling process was set to 1 to 12.Ten-fold cross-validation was performed with the training data set to determine the model hyperparameters.Model performance was evaluated by AUC with the independent test set.Results A total of 2880 machine learning models were established;of these,Mean_PCC_RFE_5_SVM had the best performance.Conclusions Machine learning models based on T2WI and ADC imaging radiomics features have significant value in differentiating early cervical cancer from chronic cervicitis.Compared with other machine learning models,the support vector machine models had higher diagnostic efficiency.

关 键 词：宫颈癌 慢性宫颈炎 放射组学特征 机器学习 

分 类 号：R445.1[医药卫生—影像医学与核医学] R736.1[医药卫生—诊断学]