基于超声影像组学及临床参数构建预测IgA肾病牛津分型T病变分级的模型  

Construction of a Machine Learning Model to Predict T Lesion Grade of Oxford Classification in IgA Nephropathy Based on Ultrasonic Radiomics and Clinical Parameters

作　　者：谌典 周畅[1] 陈伟[1] 李丽华[1] 张奥懿 聂淑婷 Chen Dian;Zhou Chang;Chen Wei;Li Lihua;Zhang Aoyi;Nie Shuting(Department of Ultrasound,Yichang Central People's Hospital,The First College of Clinical Medical Science,China Three Gorges University,Yichang 443000,China)

机构地区：[1]三峡大学第一临床医学院宜昌市中心人民医院超声科,湖北省宜昌市443000

出　　处：《中国超声医学杂志》2024年第9期1026-1030,共5页Chinese Journal of Ultrasound in Medicine

摘　　要：目的构建基于超声影像组学联合临床参数的IgA肾病(IgAN)牛津分型T病变分级的预测模型。方法收集273例经病理确诊的IgAN患者资料,以7∶3分为训练集(n=191)及验证集(n=82),根据肾小管萎缩/间质纤维化的病理将牛津分型T病变分为轻度(T0)和中重度(T1&T2)。基于常规超声图像提取组学特征,通过多种特征筛选方法构建各组学模型;纳入多因素Logistic回归分析所得独立危险因素和组学特征再次经多方法筛选,构建各联合模型,并评估各模型预测效能。结果纳入eGFR、24 h尿蛋白定量和影像组学特征,采用最优特征筛选(个数)构建的联合模型具有最佳预测效能,在训练集和验证集中的AUC分别为0.934(95%CI:0.902~0.965)、0.912(95%CI:0.853~0.971)。结论基于超声影像组学联合临床参数构建的模型可有效预测IgAN患者牛津分型的T病变分级。Objective To construct a prediction model for T lesion grade of Oxford classification in IgA nephrop-athy(IgAN)based on ultrasonic radiomics combined with clinical parameters.Methods Data of 273 patients with IgAN confirmed by pathology were retrospectively analyzed and divided into training set(n=191)and validation set(n=82)by 7:3 ratio,and T lesion of Oxford classification were graded as mild(T0)and moderate or severe(T1&T2)according to tubular atrophy/interstitial fibrosis in pathology.Radiomics features were extracted based on the conventional ultrasound images,and radiomics models were constructed by multiple feature screening methods.The independent risk factors and radiomics features obtained by multivariable Logistic regression analysis were screened again by different methods.The combination models were constructed and the predictive efficacy of each model was assessed.Results With eGFR,24h urine protein quantification and radiomics features incorporating,the combination model constructed with the optimal feature filtering(number)had the best predictive efficacy,with 0.934(95%CI:0.902-0.965),0.912(95%CI:0.853-0.971)in the training and validation sets respectively.Conclusions Model constructed based on ultrasonic radiomics and combined with clinical parameters can effectively predict T lesion grade of Oxford classification in IgAN.

关 键 词：IGA肾病 超声 影像组学 机器学习 

分 类 号：R445.1[医药卫生—影像医学与核医学] R692.31[医药卫生—诊断学]