基于机器学习影像组学的小肾肿块良恶性预测模型

A prediction model for benign and malignant small renal masses based on machine learning radiomics

摘要

目的 开发并验证一种基于增强CT的机器学习临床-影像组学综合诊断模型，从而优化SRMs的临床评估。方法 本研究为回顾性多中心研究，共纳入来自5个独立队列的349例已有病理确诊的SRMs患者，将其划分为训练队列（n=153）与独立验证队列（n=196）。采用3D-slicer软件在动脉期CT图像中勾勒ROI并提取了851个影像组学特征，通过最小绝对收缩和选择算子（LASSO）回归模型进行特征降维与筛选，将影像组学标签与独立的临床预测因子有机结合，构建出综合诊断模型。通过受试者工作特征（ROC）曲线下面积（AUC）、校准曲线、决策曲线分析（DCA）及临床影响曲线（CIC）对模型性能进行进一步评估。结果 LASSO回归最终精确筛选出29个稳健的影像组学特征。本研究构建的临床-影像组学综合模型在训练队列和验证队列中分别取得了0.907和0.935（95%CI：0.881～0.990）的AUC，性能优于传统纯临床模型（AUC 0.627，P＜0.001）。值得关注的是，单纯的影像组学模型亦展现出0.933的AUC，确立了其独立预测效能，而常规临床参数的加入仅能提供有限的增量价值。在验证集中，当最佳风险截断值（cut-off）设定为1.573时，该模型呈现了0.921的敏感度与0.778的特异度。校准曲线彰显了模型较好的拟合优度（Brier评分=0.024）。DCA与CIC分析进一步证实，在广泛的风险阈值概率区间内，该综合模型能够带来明显的临床实际净获益。结论 基于增强CT影像的无创临床-影像组学综合模型，在鉴别SRMs良恶性中表现出良好的诊断精准度与跨队列泛化能力，为辅助泌尿外科医师制定个体化的手术及随访干预策略提供了可靠的工具。

Abstract

Objective To develop and validate an integrated clinical-radiomics diagnostic model based on contrast-enhanced CT with machine learning, so as to optimize the clinical evaluation of small renal masses (SRMs). Methods A total of 349 patients with pathologically confirmed SRMs from five multi-center cohorts were retrospectively enrolled and divided into a training cohort (n=153) and an independent validation cohort (n=196). A total of 851 radiomic features were extracted from arterial phase CT images. The least absolute shrinkage and selection operator (LASSO) regression was utilized for feature selection. An integrated diagnostic model was constructed by incorporating the radiomics signature with independent clinical predictors. Model performance was assessed with the area under the receiver operating characteristic curve (AUC), calibration curves, decision curve analysis (DCA), and clinical impact curve (CIC). Results LASSO regression identified 29 robust radiomic features. The model achieved an outstanding AUC of 0.907 in the training cohort and 0.935 (95%CI: 0.881-0.990) in the validation cohort, significantly outperforming the pure clinical model (AUC=0.627, P<0.001). The optimal risk cutoff value of 1.573 yielded a sensitivity of 0.921 and a specificity of 0.778 in the validation set. Calibration curves demonstrated excellent agreement (Brier score=0.024). DCA and CIC confirmed the high clinical net benefit of the model across various threshold probabilities. Conclusion The non-invasive, clinical-radiomics integrated model based on contrast-enhanced CT demonstrated exceptional diagnostic accuracy and generalizability in differentiating benign from malignant SRMs, providing a reliable tool to facilitate personalized surgical decision-making.

关键词

小肾肿块 / 影像组学 / 诊断模型 / 机器学习

Key words

small renal masses / radiomics / diagnostic model / machine learning

参考文献

[1] SIEGEL RL, MILLER KD, FUCHS HE, et al. Cancer statistics[J]. CA Cancer J Clin, 2022, 72(1): 7-33.

[2] CLARK DJ, DHANASEKARAN SM, PETRALIA F, et al. Integrated proteogenomic characterization of clear cell renal cell carcinoma[J]. Cell 2020, 180(1): 207.

[3] ALAM R, PATEL HD, OSUMAH T, et al. Comparative effectiveness of management options for patients with small renal masses: a prospective cohort study[J]. BJU Int, 2019, 123(1): 42-50.

[4] SANCHEZ A, FELDMAN AS, HAKIMI AA. Current management of small renal masses, including patient selection, renal tumor biopsy, active surveillance, and thermal ablation[J]. J Clin Oncol, 2018, 36(36): 3591-3600.

[5] HOLLINGSWORTH JM, MILLER DC, DAIGNAULT S, et al. Rising incidence of small renal masses: a need to reassess treatment effect[J]. J Natl Cancer Inst, 2006, 98(18): 1331-1334.

[6] SUN M, THURET R, ABDOLLAH F, et al. Age-adjusted incidence, mortality, and survival rates of stage-specific renal cell carcinoma in North America: a trend analysis[J]. Eu Urol, 2011, 59(1): 135-141.

[7] ROUPRÊT M, BABJUK M, BURGER M, et al. European Association of Urology Guidelines on upper urinary tract urothelial carcinoma: 2020 update[J]. Eur Urol, 2021, 79(1): 62-79.

[8] MARCONI L, DABESTANI S, LAM TB, et al. Systematic review and meta-analysis of diagnostic accuracy of percutaneous renal tumour biopsy[J]. Eur Urol, 2016, 69(4): 660-673.

[9] MOTZER RJ, JONASCH E, BOYLE S, et al. NCCN guidelines insights: kidney cancer, version 1.2021[J]. J Natl Compr Canc Netw, 2020, 18(9): 1160-1170.

[10] KIM JH, LI S, KHANDWALA Y, et al. Association of prevalence of benign pathologic findings after partial nephrectomy with preoperative imaging patterns in the United States From 2007 to 2014[J]. JAMA Surg, 2019, 154(3): 225-231.

[11] MENON AR, PATEL V, SURESH N, et al. Avoidable benign kidney tumor resections-data from a Tertiary Care Cancer Institute[J]. J Kidney Canc VHL, 2024, 11(4): 1-9.

[12] HERRERA-CACERES JO, FINELLI A, JEWETT MAS. Renal tumor biopsy: indicators, technique, safety, accuracy results, and impact on treatment decision management[J]. World J Urol, 2019, 37(3): 437-443.

[13] LANE BR, BABINEAU D, KATTAN MW, et al. A preoperative prognostic nomogram for solid enhancing renal tumors 7 cm or less amenable to partial nephrectomy[J]. J Urol, 2007, 178(2): 429-434.

[14] KUTIKOV A, UZZO RG. The R.E.N.A.L. nephrometry score: a comprehensive standardized system for quantitating renal tumor size, location and depth[J]. J Urol, 2009, 182(3): 844-853.

[15] ZHANG J, CHEN Y, WANG X, et al. Radiologic-radiomic machine learning models for differentiation of benign and malignant solid renal masses: comparison with expert-level radiologists[J]. Am J Roentgenol, 2019, 214(1): W44-W54.

[16] COY H, HSIEH K, WU W, et al. Deep learning and radiomics: the utility of Google TensorFlow^TM Inception in classifying clear cell renal cell carcinoma and oncocytoma on multiphasic CT [J]. Abdom Radiol, 2019, 44(6): 2009-2020.

[17] PINEDA S, REAL FX, KOGEVINAS M, et al. Integration analysis of three omics data using penalized regression methods: an application to bladder cancer[J]. PLoS Genet, 2015, 11(12): e1005689.

[18] BÜHLMANN P. Proposing the vote of thanks: Regression shrinkage and selection via the Lasso: a retrospective by Robert Tibshirani[J]. J R Stat Soc B, 2010, 72(4): 417-473.

[19] FRIEDMAN J, HASTIE T, TIBSHIRANI R. Regularization paths for generalized linear models via coordinate descent[J]. J Stat Softw, 2010, 33(1): 1-22.

[20] DELONG ER, DELONG DM, CLARKE-PEARSON DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach[J]. Biometrics, 1988, 44(3): 837-845.

[21] KERR KF, BROWN MD, ZHU K, et al. Assessing the clinical impact of risk prediction models with decision curves: guidance for correct interpretation and appropriate use[J]. J Clin Oncol, 2016, 34(21): 2534-2540.

[22] VICKERS AJ, CRONIN AM. Traditional statistical methods for evaluating prediction models are uninformative as to clinical value: towards a decision analytic framework[J]. Semin Oncol, 2010, 37(1): 31-38.

[23] ASSELIN C, FINELLI A, BREAU RH, et al. Does renal tumor biopsies for small renal carcinoma increase the risk of upstaging on final surgery pathology report and the risk of recurrence?[J]. Urol Oncol, 2020, 38(10): 798.e9-798.e16.

[24] SCHIEDA N, LIM RS, MCINNES MDF, et al. Characterization of small (<4 cm) solid renal masses by computed tomography and magnetic resonance imaging: Current evidence and further development[J]. Diagn Interv Imaging, 2018, 99(7-8): 443-455.

[25] LINEHAN WM, RICKETTS CJ. The Cancer Genome Atlas of renal cell carcinoma: findings and clinical implications[J]. Nat Rev Urol, 2019, 16(9): 539-552.

[26] SUN T, TANG J, PAN YC, et al. Serum markers change for intraocular metastasis in renal cell carcinoma[J]. Biosci Rep, 2021, 41(9): BSR20203116(1-9).

[27] DE MARTINO M, KLATTE T, HAITEL A, et al. Serum cell-free DNA in renal cell carcinoma: a diagnostic and prognostic marker[J]. Cancer, 2012, 118(1): 82-90.

[28] CONTI A, DUGGENTO A, INDOVINA I, et al. Radiomics in breast cancer classification and prediction[J]. Semin Cancer Biol, 2021, 72: 238-250.

[29] LAMBIN P, LEIJENAAR RT, DEIST TM, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-762.

[30] XI IL, ZHAO Y, WANG R, et al. Deep learning to distinguish benign from malignant renal lesions based on routine MR imaging[J]. Clin Cancer Res, 2020, 26(8): 1944-1952.

[31] UHLIG J, LEHA A, DELONGE LM, et al. Radiomic features and machine learning for the discrimination of renal tumor histological subtypes: a pragmatic study using clinical-routine computed tomography[J]. Cancers (Basel), 2020, 12(10): 3010.

[32] KUNAPULI G, VARGHESE BA, GANAPATHY P, et al. A decision-support tool for renal mass classification[J]. J Digit Imaging, 2018, 31(6): 929-939.

[33] ERDIM C, YARDIMCI AH, BEKTAS CT, et al. Prediction of benign and malignant solid renal masses: machine learning-based CT texture analysis[J]. Acad Radiol, 2020, 27(10): 1422-1429.

[34] KOCAK B, DURMAZ ES, ATES E, et al. Unenhanced CT texture analysis of clear cell renal cell carcinomas: a machine learning-based study for predicting histopathologic nuclear grade[J]. AJR Am J Roentgenol, 2019, 212(6): W1-W8.

基金资助

江苏省医学重点学科（实验室）项目(ZDXK202219)

南京医科大学第一附属医院青年学者培育基金项目(PY2023040)