基于计算机断层扫描放射组学诊断肾脏良恶性囊性病变

Diagnosis of benign and malignant renal cystic lesions based on computed tomography radiomics

目的:探讨基于计算机断层扫描(CT)放射组学列线图诊断良恶性肾囊性病变(CRL)的价值。方法:分析2014年8月至2023年5月徐州医科大学附属医院Bosniak≥Ⅱ级CRL患者169例,获取每名患者临床资料、CT图像,使用Excel的RAND函数以8∶2的比例随机划分为训练集(135例)和验证集(34例),采用卡方检验、t检验等检验方法筛选相关临床危险因素。运用Pyradiomics软件从感兴趣区域(ROI)中提取大量放射组学特征,并运用最小绝对收缩与选择算子(LASSO)回归模型筛选最佳放射组学特征。结果:最大径(t训练集=-2.797,t验证集=-1.490)在恶性CRL患者中更大、Bosniak分级(χ训练集2=46.526,χ验证集2=13.852)在恶性CRL患者中更高,差异有统计学意义(P<0.05)。最终提取18个放射组学特征计算Rad-score。临床模型在训练集和验证集AUC分别为0.944[95%置信区间(confidence interval,CI):0.898~0.990]、0.855(95%CI:0.643~1.000)。放射组学模型在训练集和验证集AUC分别为0.987(95%CI:0.972~1.000)、0.910(95%CI:0.746~1.000)。联合以上两者的放射组学列线图在训练集和验证集AUC分别为0.985(95%CI:0.968~1.000)、0.931(95%CI:0.831~1.000)。结论:基于CT放射组学列线图在鉴别良恶性CRL具有良好预测价值,有助于临床医师进行准确鉴别诊断,并制定个体化治疗策略。

Objective To investigate the value of a radiomics nomogram based on computed tomography(CT)in the diagnosis of benign and malignant cystic renal lesions(CRL).Methods A total of 169 patients with CRL(Bosniak≥Ⅱ)in our hospital from August 2014 to May 2023 were enrolled in this study.Clinical data and CT images of each patient were obtained.The patients were randomly divided into the training cohort(n=135)and the validation cohort(n=34)at a ratio of 8∶2 using the RAND function of Excel.Chi-squared test,t test and other tests were used to screen relevant clinical risk factors.Pyradiomics software was used to extract a large number of radiomics features from region of interest(ROI).The least absolute shrinkage and selection operator(LASSO)regression model was used to screen the best radiomics features.Light gradient boosting machine(LightGBM)algorithm was used to construct the clinical model,radiomics model and combined model,respectively,and a nomogram combining radiomics score(Rad-score)with independent clinical factors was developed.Receiver operating characteristic(ROC)curve of each model was drawn,and the corresponding area under the ROC curve(AUC)value was calculated to measure discrimination performance of each model in training and validation cohorts.Results The maximum diameter(ttraining cohort=-2.797,tvalidation cohort=-1.490)was larger and the Bosniak classification(χtraining cohort2=46.526,χvalidation cohort2=13.852)was higher in malignant CRL patients,both showing statistically significant differences(P<0.05).A total of 18 radiomics features were finally extracted to calculate Rad-score.The clinical model performed an AUC of 0.944[95%confidence interval(CI):0.898-0.990]in the training cohort and an AUC of 0.855(95%CI:0.643-1.000)in the validation cohort.The radiomics model performed an AUC of 0.987(95%CI:0.972-1.000)in the training cohort and an AUC of 0.910(95%CI:0.746-1.000)in the validation cohort.The radiomics nomogram combining the above two models performed an AUC of 0.985(95%CI:0.972-1.000)in the training cohort and an AUC of 0.931 in the validation cohort(95%CI:0.746-1.000).Conclusion The CT-based radiomics nomogram has a good predictive value in distinguishing between BCRL and MCRL,which helps clinicians make accurate differential diagnosis and formulate individualized treatment strategies.