-
前列腺癌位居世界男性肿瘤疾病谱前列,影响着男性的生活质量[1]。我国的肿瘤监测数据[2-3]显示,前列腺癌位列我国肿瘤发病谱第7位和肿瘤死因谱第12位。前列腺癌早期临床表现不明显,多数病人就诊时已至中晚期。因此,对中晚期前列腺癌的治疗提出更高的临床需求。T3期前列腺癌属于局部晚期的一种,分为T3a期和T3b期,T3a期代表前列腺包膜切缘阳性,T3b期则表示病灶突破包膜至精囊。多参数MRI(multi-parameter MRI,mp-MRI)对前列腺疾病的诊断具有重要的参考价值[4-5],但对 < 1 mm的病灶检查效果不够理想。由于前列腺包膜与精囊在解剖上位置较接近,很难精确地在MRI图像上判断精囊是否被侵及。本研究引入影像组学技术,提取T2WI及ADC图像下T3a和T3b期前列腺癌病灶的影像组学特征,探究mp-MRI影像组学在鉴别诊断T3a、T3b期前列腺癌中的价值。
-
本研究共纳入病例151例,训练集113例,其中T3a期82例,T3b期31例;验证集38例,其中T3a期28例,T3b期10例。训练集与验证集年龄、前列腺特异抗原(PSA)水平差异均无统计学意义(P>0.05)(见表 1)。
分组 n 年龄/岁 PSA/(ng/mL) 训练集 113 69.2±6.9 26.0(24.2) 验证集 38 68.6±7.5 22.5(12.4) t — 0.45 1.05 P — >0.05 >0.05 表 1 训练集与验证集一般资料比较(x±s)
-
通过DARWIN智能科研平台在T2WI及ADC图像上共提取影像组学特征1 878个。通过最大最小值归一化、方差分析及SVM对特征进行筛选,得到最优的6个特征。将6个影像组学特征纳入组学模型,其中包括2个ADC图像特征,4个T2WI图像特征(见图 3)。
-
通过ADC图像和T2WI图像提取影像组学特征所建模型效能相对较低,训练集AUC值分别为0.71、0.79;验证集AUC值分别为0.70、0.59。通过ADC联合T2WI图像提取影像组学特征所建模型鉴别性能最优,训练集AUC值为0.79,验证集AUC值为0.73(见表 2)。使用SVM进行五折交叉验证最优模型,结果显示ADC联合T2WI图像影像组学模型校准度良好(见图 4)。
模型 训练集 验证集 AUC(95%CI) 敏感性/% 特异性/% 准确性/% AUC(95%CI) 敏感性/% 特异性/% 准确性/% ADC 0.71(0.55~0.87) 61.3 80.5 78.8 0.70(0.40~0.96) 90.0 60.7 68.4 T2WI 0.79(0.66~0.92) 80.6 65.9 72.6 0.59(0.30~0.89) 90.0 53.6 73.7 ADC+T2WI 0.79(0.64~0.95) 61.3 96.3 81.4 0.73(0.47~1.00) 80.0 67.9 76.3 表 2 3种模型在训练集和验证集中的统计学参数
多参数MRI影像组学在鉴别诊断T3a与T3b期前列腺癌中的价值
Value of multi-parameter MRI-based radiomics in distinguishing T3a stage from T3b stage of prostate cancer
-
摘要:
目的探讨多参数MRI(multi-parameter MRI, mp-MRI)影像组学在鉴别诊断T3a、T3b期前列腺癌中的价值。 方法回顾性分析151例T3期前列腺癌病人的影像资料, 其中T3a期110例, T3b期41例。使用达尔文智能科研平台在T2WI及ADC横轴位图像上手动勾画感兴趣区, 提取影像组学特征。按照3:1的比例分别将ADC图像、T2WI图像、ADC图像联合T2WI图像三种模式下的影像组学特征分为训练集和验证集, 依次对3组训练集构建模型, 使用相应验证集进行内部验证。通过ROC曲线对模型进行分析, 并评价不同序列的诊断效能。 结果通过达尔文科研平台共提取1878个影像组学特征, 最终保留6个(ADC相关2个, T2WI相关4个)(P < 0.05), 单独T2WI和ADC图像下训练集AUC值分别为0.79、0.71, 验证集AUC值为0.59、0.70;ADC图像联合T2WI图像下训练集AUC值为0.79, 验证集AUC值为0.73。 结论T2WI图像联合ADC图像影像组学对术前鉴别T3a、T3b期前列腺癌具有较好诊断价值, 可在一定程度上弥补MRI对 < 1 mm病灶检查效果欠佳的缺陷, 为癌灶是否侵犯精囊提供补充, 协助临床术前获得更加详实的资料, 为精准手术提供指导。 Abstract:ObjectiveTo investigate the value of radiomic features of multi-parameter MRI(mp-MRI) for distinguishing T3a stage from T3b stage of prostate cancer. MethodsThe imaging data of 151 patients with T3 stage prostate cancer were retrospectively analyzed, including 110 patients with T3a stage and 41 patients with T3b stage.Using DARWIN research platform, regions of interest(ROI) were manually sketched on the transverse T2WI and ADC images to extract the radiomic features of image.The imaging data of ADC, T2WI and ADC combined with T2WI were divided into training group and verification group at the ratio of 3:1.Then the model of three groups of training group was built, and the corresponding tests were used for internal verification.The diagnostic efficacy of different sequences was analyzed by the ROC curve. ResultsA total of 1878 imaging features were extracted by DARWIN research platform, and 6 features(2 features of ADC and 4 features of T2WI) were retained(P < 0.05).The AUC values of training group and validation group under simple T2WI or ADC sequence were 0.79, 0.71, 0.59, 0.70.The AUC value of training group under ADC combined with T2WI sequence was 0.79, and that of validation group was 0.73. ConclusionsT2WI combined with ADC imaging radiomics has a good auxiliary value in the preoperative differential diagnosis of T3a and T3b prostate cancer, which can make up for the defect of MRI in the diagnosis of lesions less than 1 mm and provide the supplement for the invasion of seminal vesicle to obtain more clinical preoperative detailed information and provide the guidance for accurate surgery. -
Key words:
- prostate neoplasms /
- radiomics /
- multi-parameter MRI /
- T3a stage /
- T3b stage
-
表 1 训练集与验证集一般资料比较(x±s)
分组 n 年龄/岁 PSA/(ng/mL) 训练集 113 69.2±6.9 26.0(24.2) 验证集 38 68.6±7.5 22.5(12.4) t — 0.45 1.05 P — >0.05 >0.05 表 2 3种模型在训练集和验证集中的统计学参数
模型 训练集 验证集 AUC(95%CI) 敏感性/% 特异性/% 准确性/% AUC(95%CI) 敏感性/% 特异性/% 准确性/% ADC 0.71(0.55~0.87) 61.3 80.5 78.8 0.70(0.40~0.96) 90.0 60.7 68.4 T2WI 0.79(0.66~0.92) 80.6 65.9 72.6 0.59(0.30~0.89) 90.0 53.6 73.7 ADC+T2WI 0.79(0.64~0.95) 61.3 96.3 81.4 0.73(0.47~1.00) 80.0 67.9 76.3 -
[1] SIEGEL RL, MILLER KD, JEMAL A. Cancer statistics, 2017[J]. CA Cancer J Clin, 2017, 65(1): 5. [2] JAGAI JS, MESSER LC, RAPPAZZO KM, et al. County-level cumulative environmental quality associated with cancer incidence[J]. Cancer, 2017, 123(15): 2901. doi: 10.1002/cncr.30709 [3] CHEN W, ZHENG R, BAADE PD, et al. Cancer statistics in China, 2015[J]. CA Cancer J Clin, 2016, 66(2): 115. doi: 10.3322/caac.21338 [4] BJURLIN MA, MENDHIRATTA N, TANEJA SS. Multiparametric MRI of the prostate as a tool for prostate cancer detection, localization, and risk assessment[J]. Cent Eur J Urol, 2016, 69(1): 9. doi: 10.1016/j.eururo.2015.07.004 [5] 汪维, 张青, 张冰, 等. 自由臂经会阴多参数磁共振-超声影像融合引导靶向穿刺诊断前列腺癌的前瞻性研究[J]. 中华泌尿外科杂志, 2018, 39(3): 192. doi: 10.3760/cma.j.issn.1000-6702.2018.03.009 [6] WEINREB JC, BARENTSZ JO, CHOYKE PL, et al. PI-RADS prostate imaging reporting and data system: 2015, Version 2[J]. Eur Urol, 2016, 69: 16. doi: 10.1016/j.eururo.2015.08.052 [7] 韩苏军, 张思维, 陈万青, 等. 中国前列腺癌发病现状和流行趋势分析[J]. 临床肿瘤学杂志, 2013, 18(4): 330. doi: 10.3969/j.issn.1009-0460.2013.04.009 [8] 熊升远, 阮同德. T3期前列腺癌治疗进展[J]. 国际泌尿系统杂志, 2015, 35(2): 279. doi: 10.3760/cma.j.issn.1673-4416.2015.02.038 [9] 罗超. T3期前列腺癌治疗的进展[J]. 世界最新医学信息文摘, 2019, 19(55): 36. [10] 李鑫, 王丽丽, 汪浩, 等. 去势抵抗性前列腺癌新型内分泌药物治疗次序的研究进展[J]. 肿瘤, 2017, 37(9): 995. [11] SCHULMAN C, CORNEL E, MATVEEV V, et al. Intermittent versus continuous androgen deprivation therapy in patients with relapsing or locally advanced prostate cancer: a phase 3b randomised study[J]. Eur Urol, 2016, 69(4): 720. doi: 10.1016/j.eururo.2015.10.007 [12] MAKAREWICZ R, ROSZKOWSKI K, LEBIODA A, et al. PSA bounces after brachytherapy HDR and external beam radiation therapy for prostate cancer[J]. Rep Pract Oncol Radioth, 2006, 11(5): 217. doi: 10.1016/S1507-1367(06)71066-5 [13] 陆黎, 孙宗琼, 李超凡, 等. 对比MRI与CT诊断不同病理分期前列腺癌的准确率[J]. 中国性科学, 2018, 27(12): 12. doi: 10.3969/j.issn.1672-1993.2018.12.003 [14] POLANEC SH, BICKEL H, WENGERT GJ, et al. Can the addition of clinical information improve the accuracy of PI-RADS version2 for the diagnosis of clinically significant prostate cancer in positive MRI?[J]. Clin Radiol, 2020, 75(2): 1. [15] XU M, FANG M, ZOU J, et al. Using biparametric MRI radiomics signature to differentiate between benign and malignant prostate lesions[J]. Eur J Radiol, 2019, 114(5): 38. [16] 张永生, 刘海明, 叶裕丰, 等. MR动态增强、DWI/ADC值等多参数成像与前列腺癌Gleason分级相关性研究[J]. 国际泌尿系统杂志, 2017, 37(6): 842. doi: 10.3760/cma.j.issn.1673-4416.2017.06.013