摘要
目的 基于深度神经网络(DNN)构建超声弹性成像鉴别宫颈占位性病变(SOL)良恶性的智能诊断模型并分析其效能。方法 回顾性分析2018年1月—2022年12月、2023年1月—2024年1月本院收治的266例和115例宫颈SOL患者的临床资料,分别作为训练组、验证组。所有患者入院后未治疗前均行超声弹性成像检查。收集训练组超声弹性成像图像并在预处理后输入计算机数据库,通过预训练获取图像特征,进行图像特征分类并输出,构建基于DNN的智能诊断模型。对验证组,以组织病理学结果为金标准判断基于DNN的智能诊断模型、超声弹性成像对宫颈SOL良恶性的诊断灵敏度、特异度及准确度,采用Kappa检验评估基于DNN的智能诊断模型、超声弹性成像诊断结果与病理结果的一致性。结果 验证组115例患者中,组织病理检查证实恶性60例,良性55例;基于DNN的智能诊断模型诊断验证组的灵敏度、特异度及准确度分别为95.00%、96.36%、95.65%,高于超声弹性成像的86.67%、89.09%、87.83%;Kappa检验结果显示,基于DNN的智能诊断模型诊断验证组的结果与病理学结果一致性高(P<0.001),超声弹性成像诊断验证组的结果与病理学结果存在一致性(P<0.001)。结论 本研究成功构建了基于DNN的智能诊断模型,并且验证了其可提升宫颈SOL良恶性的鉴别诊断效能。
Objective To construct an intelligent diagnosis model based on deep neural networks(DNN)for differentiating benign and malignant cervical space occupying lesions(SOL)by ultrasonic elastic imaging and analyze its efficacy.Methods The clinical data of 266 patients and 115 patients suffering from cervical SOL who were admitted to our hospital from January 2018 to December 2022,from January 2023 to January 2024,were retrospectively analyzed,and they were enrolled in the training group and the verification group,respectively.All patients underwent ultrasound elastography examination after admission and before treatment.The ultrasonic elastic imaging images of the training group were collected and input into the computer database after pre-processing.Image features were obtained through pre-training,image features were classified and output,and an intelligent diagnosis model based on DNN was constructed.For the verification group,histopathological results were used as the gold standard to judge the diagnostic sensitivity,specificity and accuracy of intelligent diagnosis model based on DNN and ultrasonic elastic imaging for benign and malignant cervical SOL.Kappa test was used to evaluate the consistency of intelligent diagnosis model based on DNN and ultrasonic elastic imaging diagnosis results with pathological results.Results Among the 115 patients in the verification group,60 cases were confirmed to be malignant by histopathology examination and 55 cases were benign.The sensitivity,specificity and accuracy of the intelligent diagnostic model based on DNN were 95.00%,96.36%and 95.65%,respectively,which were higher than 86.67%,89.09%and 87.83%of ultrasonic elastic imaging.The results of Kappa test showed that the diagnostic results of the intelligent diagnostic model based on DNN in verification group were highly consistent with the results of pathology(P<0.001),and the diagnostic results of the ultrasound elastography in validation group were consistent with the pathological results(P<0.001).Conclusions This study su
作者
高姗姗
付梦真
王永莉
Gao Shanshan;Fu Mengzhen;Wang Yongli(Ultrasonography Lab of Inpatient Department,Luohe Central Hospital,Luohe,Henan 462000,China.)
出处
《齐齐哈尔医学院学报》
2024年第17期1663-1667,共5页
Journal of Qiqihar Medical University
关键词
宫颈占位性病变
良性
恶性
超声弹性成像
深度神经网络
Cervical space occupying lesion
Benign
Malignant
Ultrasonic elastography
Deep neural network
