基于超高分辨和常规超声造影的机器学习模型对淋巴结结核的诊断效能比较  

The comparison of the diagnostic efficacy of machine learning models based on ultra-high resolution contrast-enhanced ultrasound and conventional contrast-enhanced ultrasound

作　　者：杨高怡 王莹[1] 张莹 陈佩君[2] 童嘉辉 俞跃辉 林婷 颜心怡[3] 罗佳磊 Yang Gaoyi

机构地区：[1]杭州市第一人民医院,310000 [2]杭州市红十字会医院,310000 [3]杭州师范大学,310000 [4]浙江中医药大学,310053

出　　处：《浙江临床医学》2024年第10期1426-1429,F0004,共5页Zhejiang Clinical Medical Journal

基　　金：浙江省医药卫生科技计划项目(2021KY911);浙江省医药卫生技术项目(2024KY1231);杭州市卫健委重大项目(Z20230098)。

摘　　要：目的探讨在机器学习帮助下超高分辨以及普通超声造影对淋巴结结核的诊断价值。方法前瞻性收集2021年1月至2024年1月于杭州市红十字会医院就诊的颈部淋巴结肿大患者198例,并按7∶3比例随机分为训练集和验证集,通过机器学习的方法分别建立常规超声造影(Normal CEUS)模型以及高分辨超声造影(HR CEUS)模型,比较并分析两个模型的诊断效能。结果Normal CEUS模型在训练集以及验证集中的AUC分别为0.820和0.798。HR CEUS模型在训练集以及验证集中的AUC(0.993和0.990)高于Normal CEUS模型,其在验证集中的特异度(100%)也高于Normal CEUS模型的特异度(60.9%)。结论基于机器学习的超高分辨超声造影模型比常规模型更具有诊断价值。Objective To explore the diagnostic value of ultra-high resolution contrast-enhanced ultrasound(UHRUS)and conventional contrast-enhanced ultrasound(CEUS)in lymph node tuberculosis with the help of machine learning.Methods Prospective collection of 198 patients with cervical lymphadenopathy who visited Hangzhou Red Cross Hospital from January 2021 to January 2024,and randomly divided them into a training set and a validation set in a 7:3 ratio.Normal CEUS model and HR CEUS model were established using machine learning methods,and the diagnostic efficacy of the two models was compared and analyzed.Results The area under the curve(AUC)of the normal CEUS model in the training set and validation set are 0.820 and 0.798,respectively.The AUC(0.993 and 0.990)of the HR CEUS model in the training and validation sets were higher than those of the normal CEUS model,and its specificity(100%)in the validation set was also higher than that of the normal CEUS model(60.9%).Conclusion The ultra-high resolution CEUS model based on machine learning has better diagnostic value than the conventional CEUS model.

关 键 词：超高分辨超声 超声造影 机器学习 淋巴结结核 

分 类 号：R445.1[医药卫生—影像医学与核医学] R522[医药卫生—诊断学]