-
前列腺癌(prostate cancer,PCa)是男性泌尿系统肿瘤中最常见的恶性肿瘤之一。据2018年全球癌症统计报告显示,新增PCa病人占男性癌症新发的13.5%[1]。早期发现和诊断PCa有利于病人的预后。临床中常用的直肠指检仅能初步判定前列腺大小及表面质地情况;前列腺特异性抗原(prostate-specific antigen,PSA)检测特异性不强,其在前列腺增生、前列腺炎中也有升高[2];经直肠超声引导活检则不良反应较大,多参数磁共振成像作为一种非
侵袭性的方法越来越受欢迎[3]。磁共振成像的软组织分辨力高,并且可以将形态学和功能成像结合起来从而提高PCa的诊断精确性[4]。PCa的风险评估对指导治疗方式具有重要意义,磁共振影像组学是借助计算机辅助诊断技术利用特征分析将传统医学影像转化为大量可挖掘的数据特征,通过机器学习算法建立分类模型,可以无创评估PCa进展风险。基于此,本研究对T2WI和ADC图的影像组学模型能否鉴别出高危和中低危PCa进行分析。现作报道。
-
训练队列病人182例(高危PCa 106例,中低危PCa 76例),验证队列83例(高危PCa 49例,中低危PCa 34例)。训练队列和验证队列病人年龄、PSA差异均无统计学意义(P>0.05)(见表 1)。
分组 n 年龄(x±s)/岁 PSA/(ng/mL) 高危组 训练队列 106 69.3±6.9 14.5(8.1,29.9) 验证队列 49 68.9±7.3 13.0(9.8,28.6) t — 0.33 0.24* P — >0.05 >0.05 中低危组 训练队列 76 68.4±7.0 9.5(6.9,15.1) 验证队列 34 66.2±8.3 8.5(5.7,14.1) t — 1.73 0.92* P — >0.05 >0.05 *示秩和检验 表 1 训练队列与验证队列病人一般资料比较[M(P25,P75)]
-
本研究提取高危和中低危PCa病人ADC图和T2WI序列各939个特征,利用LASSO回归模型最后得到10个影像组学特征用于构建组学模型,其中包括6个ADC特征,4个T2WI特征(见图 2)。
-
影像组学模型仅使用T2WI特征预测效能一般,训练队列ROC曲线下面积(AUC)为0.70(95%CI 0.63~0.77),验证队列为0.58(95%CI 0.47~0.68)。使用ADC图像预测效能较好,训练队列AUC为0.79(95%CI 0.72~0.85),验证队列AUC为0.78(95%CI 0.68~0.86)(见图 3)。T2WI联合ADC图构建的影像组学模型表现出最优的预测效能,训练队列AUC为0.84(95%CI 0.78~0.89),验证队列AUC为0.80(95%CI 0.69~0.88)(见图 4)。
磁共振影像组学在鉴别中低危和高危前列腺癌中的应用
Application of magnetic resonance imaging in differentiating low-and medium-risk prostate cancer from high-risk prostate cancer
-
摘要:
目的建立基于支持向量机学习算法的影像组学模型,研究其鉴别高危前列腺癌与中低危前列腺癌的诊断效能。 方法回顾性分析265例经病理证实的前列腺癌病人,其中高危病人155例,中低危病人110例。所有病人术前均进行MRI检查。由两位放射医师使用达尔文智能科研平台手动勾画感兴趣区,从每例病人的T2WI和ADC图中分别提取影像组学特征,采用受试者工作特征(ROC)曲线及ROC曲线下面积(AUC)验证影像组学特征的鉴别效能,对比T2WI、ADC及T2WI+ADC的诊断价值。 结果共筛选出10个影像组学特征(6个ADC序列特征,4个T2WI序列特征)可以用来鉴别高危及中低危前列腺癌。仅使用T2WI获得的组学模型鉴别效能较低,训练队列AUC为0.70(95%CI 0.63~0.77),验证队列AUC为0.58(95%CI 0.47~0.68)。ADC图组学模型预测效能较好,训练队列AUC为0.79(95%CI 0.72~0.85),验证队列AUC为0.78(95%CI 0.68~0.86)。T2WI联合ADC图构建的影像组学模型表现出最优预测效能,训练队列AUC为0.84(95%CI 0.78~0.89),验证队列AUC为0.80(95%CI 0.69~0.88)。 结论本研究构建的基于T2WI和ADC图的影像组学模型在一定程度上对中低危及高危前列腺癌病人进行区分,为前列腺癌的分期提供了一种无创的预测方式,指导治疗方案的选择。 Abstract:ObjectiveTo establish a radiomic model based on SVM learning algorithm to evaluate the diagnostic efficiency in differentiating high-risk prostate cancer(PCa) from low-and medium-risk PCa. Methods265 patients with PCa confirmed by histopathologic results were analyzed retrospectively, including 155 high risk patients and 110 low-and medium-risk patients.All patients were examined by MRI before operation.Regions of interest(ROIs)were manually delineated by two radiologists using DARWIN research platform, and the radiomic features were extracted from each segmented ROI of the T2WI and ADC images.The Receiver Operating Characteristic(ROC) curve and the area under the ROC curve(AUC) were used to validate the differential diagnosticefficiency of radiomic features, and the diagnostic performance of T2WI, ADC and T2WI+ADC were compared. ResultsA total of 10 radiomic features were selected to identify high-risk, low-and medium-risk prostate cancer.The classification performance of the T2WI-based radiomic model was not satisfying with an AUC of 0.70(95%CI 0.63-0.77) in the training set and 0.58(95%CI 0.47-0.68) in the validation set.ADC-based model performed better with the training set achieved AUC of 0.79(95%CI 0.72-0.85) and 0.78(95%CI 0.68-0.86) of the validation set.The ensemble model constructed by both T2WI and ADC achieved the highest predictive AUCs, 0.84(95%CI 0.78-0.89) for training set and 0.80(95%CI 0.69-0.88) of the validation. ConclusionsThe radiomic model based on T2WI and ADC maps distinguished patients with different levels of prostate cancer risk to a certain extent, which provides a non-invasive prediction method for the classification and treatment guidance of Pca. -
Key words:
- prostate cancer /
- radiomics /
- magnetic resonance imaging /
- risk stratification /
- support vector machine
-
表 1 训练队列与验证队列病人一般资料比较[M(P25,P75)]
分组 n 年龄(x±s)/岁 PSA/(ng/mL) 高危组 训练队列 106 69.3±6.9 14.5(8.1,29.9) 验证队列 49 68.9±7.3 13.0(9.8,28.6) t — 0.33 0.24* P — >0.05 >0.05 中低危组 训练队列 76 68.4±7.0 9.5(6.9,15.1) 验证队列 34 66.2±8.3 8.5(5.7,14.1) t — 1.73 0.92* P — >0.05 >0.05 *示秩和检验 -
[1] BRAY F, FERLAY J, SOERJOMATARAM I, et al. A Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2018, 68(6): 394. doi: 10.3322/caac.21492 [2] GEORGE AK, TURKBEY B, VALAYIL SG, et al. A urologist's perspective on prostate cancer imaging: past, present, and future[J]. Abdom Radiol(NY), 2016, 41(5): 805. doi: 10.1007/s00261-016-0751-6 [3] OBERLIN DT, CASALINO DD, MILLER FH, et al. Dramatic increase in the utilization of multiparametric magnetic resonance imaging for detection and management of prostate cancer[J]. Abdom Radiol(NY), 2017, 42(4): 1255. doi: 10.1007/s00261-016-0975-5 [4] WANG S, BURTT K, TURKBEY B, et al. Computer aided-diagnosis of prostate cancer on multiparametric MRI: a technical review of current research[J]. Biomed Res Int, 2014, 2014: 789561. [5] MOHLER JL, ANTONARAKIS ES, ARMSTRONG AJ, et al. Prostate Cancer, Version 2.2019, NCCN Clinical Practice Guidelines in Oncology[J]. J Natl Compr Canc Netw, 2019, 17(5): 479. doi: 10.6004/jnccn.2019.0023 [6] MORIS L, CUMBERBATCH MG, VAN DEN BROECK T, et al. Benefits and risks of primary treatments for high-risk localized and locally advanced prostate cancer: an international multidisciplinary systematic Review[J]. Eur Urol, 2020, 77(5): 614. doi: 10.1016/j.eururo.2020.01.033 [7] MOSCHINI M, CARROLL PR, EGGENER SE, et al. Parker C. Low-risk prostate cancer: identification, management, and outcomes[J]. Eur Urol, 2017, 72(2): 238. doi: 10.1016/j.eururo.2017.03.009 [8] SOSNOWSKI R, KAMECKI H, DANESHMAND S, et al. Active surveillance for low-risk prostate cancer-in pursuit of a standardized protocol[J]. Cent European J Urol, 2020, 73(2): 123. [9] TOSOIAN JJ, MAMAWALA M, EPSTEIN JI, et al. Active surveillance of grade group 1 prostate cancer: long-term outcomes from a large prospective cohort[J]. Eur Urol, 2020, 77(6): 675. doi: 10.1016/j.eururo.2019.12.017 [10] HAMDY FC, DONOVAN JL, LANE JA, et al. 10-year outcomes after monitoring, surgery, or radiotherapy for localized prostate cancer[J]. N Engl J Med, 2016, 375(15): 1415. doi: 10.1056/NEJMoa1606220 [11] BUTLER SS, MAHAL BA, LAMBA N, et al. Use and early mortality outcomes of active surveillance in patients with intermediate-risk prostate cancer[J]. Cancer, 2019, 125(18): 3164. doi: 10.1002/cncr.32202 [12] JANG TL, PATEL N, FAIENA I, et al. Comparative effectiveness of radical prostatectomy with adjuvant radiotherapy versus radiotherapy plus androgen deprivation therapy for men with advanced prostate cancer[J]. Cancer, 2018, 124(20): 4010. doi: 10.1002/cncr.31726 [13] GONG L, XU M, FANG M, et al. Noninvasive prediction of high-grade prostate cancer via biparametric MRI radiomics[J]. J Magn Reson Imaging, 2020, 52(4): 1102. doi: 10.1002/jmri.27132 [14] VIGNATI A, MAZZETTI S, GIANNINI V, et al. Texture features on T2-weighted magnetic resonance imaging: new potential biomarkers for prostate cancer aggressiveness[J]. Phys Med Biol, 2015, 60(7): 2685. doi: 10.1088/0031-9155/60/7/2685 [15] ALGOHARY A, SHIRADKAR R, PAHWA S, et al. Combination of peri-tumoral and intra-tumoral radiomic features on bi-parametric MRI accurately stratifies prostate cancer risk: a multi-site study[J]. Cancers(Basel), 2020, 12(8): 2200. [16] FEHR D, VEERARAGHAVAN H, WIBMER A, et al. Automatic classification of prostate cancer Gleason scores from multiparametric magnetic resonance images[J]. Proc Natl Acad Sci U S A, 2015, 112(46): E6265. doi: 10.1073/pnas.1505935112 [17] LOSNEGÅRD A, REISÆTER LAR, HALVORSEN OJ, et al. Magnetic resonance radiomics for prediction of extraprostatic extension in non-favorable intermediate- and high-risk prostate cancer patients[J]. Acta Radiol, 2020, 61(11): 1570. doi: 10.1177/0284185120905066 [18] SAHRAN S, ALBASHISH D, ABDULLAH A, et al. Absolute cosine-based SVM-RFE feature selection method for prostate histopathological grading[J]. Artif Intell Med, 2018, 87: 78. doi: 10.1016/j.artmed.2018.04.002 [19] PATEL N, HENRY A, SCARSBROOK A. The value of MR textural analysis in prostate cancer[J]. Clin Radiol, 2019, 74(11): 876. doi: 10.1016/j.crad.2018.11.007