分享:
分享到微信朋友圈
X
临床研究
基于扩散模型生成的高b值DWI评估前列腺癌根治性治疗后局部复发的应用价值
邓文友 郭小芳 胡奎 胡磊

Cite this article as: DENG W Y, GUO X F, HU K, et al. Application of high b-value DWI generated based on diffusion model to assess local recurrence after radical treatment of prostate cancer[J]. Chin J Magn Reson Imaging, 2024, 15(9): 86-93.本文引用格式:邓文友, 郭小芳, 胡奎, 等. 基于扩散模型生成的高b值DWI评估前列腺癌根治性治疗后局部复发的应用价值[J]. 磁共振成像, 2024, 15(9): 86-93. DOI:10.12015/issn.1674-8034.2024.09.015.


[摘要] 目的 探讨基于扩散模型生成高b值扩散加权成像(diffusion weighted imaging, DWI)对前列腺癌根治性治疗后局部复发的评估价值。材料与方法 回顾性分析63例前列腺癌行根治性放射治疗(radiation therapy, RT)或根治性前列腺切除术(radical prostatectomy, RP)后出现生化复发(biochemical recurrence, BCR)患者的临床及影像相关资料,其中RT组21例,RP组42例。将患者初始表观扩散系数(apparent diffusion coefficient, ADC)图经过计算得到的DWI图像输入前列腺DWI生成模型,获得生成的高b值(b=2000 s/mm2)DWI图。通过3位阅片者对计算DWI及生成DWI进行图像质量评价,并根据前列腺复发影像报告(Prostate Imaging for Recurrence Reporting, PI-RR)系统对所有病例进行复发风险评分采用多读者多病例受试者工作特征(multi-reader multi-case receiver operating characteristic, MRMC-ROC)曲线比较不同阅片者的诊断效能差异。等级评分一致性采用组内相关系数进行检验。结果 3位阅片者对生成DWI组图像质量评价均优于计算DWI组(P=0.002、0.003、0.002)。3位阅片者对RT组的生成DWI与计算DWI组的PI-RR总评分差异有统计学意义(P=0.031、0.049、0.041);3位阅片者对RP组生成DWI与计算DWI组的PI-RR总评分差异有统计学意义(P=0.034、0.049、0.036)。3位阅片者使用生成DWI进行RT与RP两组PI-RR总评分预测发生局部复发的曲线下面积(area under the curve, AUC)值范围分别为0.884~0.924、0.926~0.947;利用计算DWI进行RT与RP两组PI-RR总评分预测发生局部复发的AUC值范围分别为0.783~0.792、0.843~0.893。合并RT及RP两组病例后,使用PI-RR总评分预测所有患者局部复发的状态,生成DWI组与计算DWI组的AUC值范围分别为0.912~0.930、0.797~0.858。结论 基于扩散模型生成的高b值DWI能显著提高前列腺癌根治性治疗后局部复发诊断效能。
[Abstract] Objective To investigate the value of generating high b-value diffusion weighted imaging (DWI) based on diffusion model for the assessment of local recurrence after radical treatment of prostate cancer.Materials and Methods Retrospective analysis of the clinical and imaging data of 63 patients with biochemical recurrence (BCR) after radical radiotherapy (RT) or radical prostatectomy (RP) for prostate cancer, including 21 patients in the RT group and 42 patients in the RP group. DWI images calculated using the patient's initial apparent diffusion coefficient (ADC) maps were input into the prostate DWI generated model to obtain the generated high b-value (b=2000 s/mm2) DWI maps. The image quality of the calculated DWI and the generated DWI was evaluated by 3 readers, and the risk of recurrence was scored in all cases according to the Prostate Imaging for Recurrence Reporting (PI-RR) system score. Multi-reader multi-case receiver operating characteristic (MRMC-ROC) curve were used to compare the differences in diagnostic efficacy between different readers. Grade score agreement was tested using intragroup correlation coefficients.Results All three readers rated the image quality of the generated DWI group better than that of the calculated DWI group (P=0.002, 0.003, 0.002). The difference in the total PI-RR scores between the generated DWI and calculated DWI groups of the RT group by the three readers was statistically significant (P=0.031, 0.049, 0.041). The difference in PI-RR total score between the generated DWI and calculated DWI groups was statistically significant (P=0.034, 0.049, 0.036). The range of area under the curve (AUC) values for PI-RR total score prediction of the occurrence of localized recurrence in the RT and RP groups by the three readers using the generated DWI was categorized as 0.884-0.924 and 0.926-0.947; the range of AUC value for PI- RR total score prediction of the occurrence of local recurrence in the RT and RP groups by the three readers using the calculated DWI was categorized as 0.783-0.792 and 0.843-0.893. After combining the cases in RT and RP groups, the PI-RR total score was used to predict the status of local recurrence in all the patients, and the range of AUC values for the generated DWI group and the calculated DWI group were 0.912-0.930 and 0.797-0.858.Conclusions High b-value DWI generated based on the diffusion model can significantly improve the diagnostic efficacy of local recurrence after radical treatment of prostate cancer.
[关键词] 前列腺癌;磁共振成像;扩散模型;扩散加权成像;根治性治疗;局部复发
[Keywords] prostate cancer;magnetic resonance imaging;diffusion model;diffusion weighted imaging;radical treatment;local recurrence

邓文友 1   郭小芳 1   胡奎 1   胡磊 2*  

1 湖北省肿瘤医院放射科,武汉 430079

2 广东省人民医院放射科,广州 510080

通信作者:胡磊,E-mail: hulei@gdph.org.cn

作者贡献声明::胡磊设计本研究的方案,对稿件重要内容进行了修改,获得了国家自然科学基金资助;邓文友起草和撰写稿件,获取、分析和解释本研究的数据;郭小芳、胡奎获取、分析和解释本研究的数据,对稿件重要内容进行了修改;全体作者都同意发表最后的修改稿,同意对本研究的所有方面负责,确保本研究的准确性和诚信。


基金项目: 国家自然科学基金项目 82302130
收稿日期:2024-05-15
接受日期:2024-08-09
中图分类号:R445.2  R737.25 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2024.09.015
本文引用格式:邓文友, 郭小芳, 胡奎, 等. 基于扩散模型生成的高b值DWI评估前列腺癌根治性治疗后局部复发的应用价值[J]. 磁共振成像, 2024, 15(9): 86-93. DOI:10.12015/issn.1674-8034.2024.09.015.

0 引言

       早期前列腺癌根治性治疗方法主要包括根治性前列腺切除术(radical prostatectomy, RP)和根治性放射治疗(radiation therapy, RT)[1],但治疗后仍会有部分患者出现生化复发(biochemical recurrence, BCR)。有研究证据表明,患者出现BCR后发生肿瘤复发及转移的风险存在较大差异[2]。27%~53%的前列腺癌患者在RP后会发生BCR,然而,只有16%的BCR患者最终会死于前列腺癌复发[3]。相对于BCR,局部复发的检测对于预测患者术后风险意义更大[4]。然而,目前临床医生在制订前列腺根治性切除术后复发治疗方案时主要以BCR为参考,并不严格区分BCR与局部复发的真实状态[2, 5, 6]。2022年国内外前列腺癌诊疗指南《EAU-EANM-ESTRO-ESUR-ISUP-SIOG Guidelines on Prostate Cancer》以及《2022版中国泌尿外科和男科疾病诊断治疗指南》均指出对RP后BCR患者应该进行更为精准的风险分层,明确BCR患者是否发生局部复发,能够避免低风险患者的过度治疗和高风险患者治疗不足,具有十分重要的临床价值[7]。2021年,国际专家组发布了前列腺复发报告成像(Prostate Imaging for Recurrence Reporting, PI-RR)系统,旨在指导前列腺癌根治性治疗后进行多参数MRI(multiparametric MRI, mpMRI)检查、解读和报告评估局部复发情况[8, 9, 10]。该系统指出高b值(b≥1400 s/mm2)扩散加权成像(diffusion weighted imaging, DWI)有助于提升局部复发的检测准确度[11]。但是,b值的增加通常伴随着DWI图像质量的下降。同时受扫描仪器、扫描序列、患者自身条件的影响,扫描得到的高b值DWI图像通常面临着平面空间分辨率不足、失真和伪影等问题[12]。另一种获得高b值DWI的方法是通过拟合的信号衰减曲线来计算高b值DWI图像[13]。该方法一定程度减少了高b值DWI的扫描时间,提升了图像质量,但是,计算高b值DWI依然依赖中低b值DWI的图像质量。由于RP或RT后,瘤床或术区的正常解剖结构也发生了变化,传统的计算高b值DWI的图像质量更难保证,增加了利用mpMRI评估患者是否局部复发的难度,因此,对于获取高质量的高b值DWI图像在前列腺癌治疗后评估局部复发的工作显得尤为重要。

       近年来,医学影像深度学习算法的崛起为解决诸多传统的医学图像问题提供了新思路[14, 15, 16]。基于生成对抗网络、深度学习重建及级联学习等方法已经被诸多研究者应用到前列腺MRI图像的降噪、高b值高分辨重建、去伪影、跨模态生成等诸多任务当中[17, 18, 19, 20]。既往的研究更多关注前列腺病变治疗前的DWI图像优化处理及癌症风险评估,目前关于提高前列腺癌根治性治疗后的高b值DWI图像质量的相关研究内容尚待完善。本研究基于扩散模型利用深度学习算法建立生成模型来提升前列腺癌RP或RT后复查MRI采集的高b值DWI图像质量,旨在利用生成的高质量DWI图像来帮助放射科医生更好地评估前列腺治疗后局部复发的状态。

1 材料与方法

1.1 研究对象

       本研究遵守《赫尔辛基宣言》,经湖北省肿瘤医院伦理机构审查委员会批准,免除受试者知情同意,批准文号:LLHBCH2023YN-052。回顾性收集2017年5月至2022年7月经湖北省肿瘤医院泌尿外科及放射治疗中心收治的前列腺癌患者共298例,所有患者在治疗前均完成了前列腺MRI检查,在根治性治疗结束后随访期间出现BCR共82例。纳入标准:(1)遵循RP或RT后BCR和前列腺特异性抗原(prostate-specific antigen, PSA)持续升高(其中RP后连续两次检测PSA>0.2 ng/mL,RT后检测PSA较最低值升高2 ng/mL)[2];(2)出现BCR后完成前列腺MRI检查,且全身发射型计算机断层成像(emission computerized tomography, ECT)扫描为阴性。排除标准:(1)一般临床资料及相关影像随访数据缺失;(2)缺乏相应前列腺MRI扫描的必要序列。

       局部复发阳性结果定义:(1)前列腺或前列腺切除术床病理活检结果阳性[21];(2)超过24个月随访既往影像学检查发现的复发灶治疗后缩小或PSA水平降低[10];(3)随访影像检查[包括盆腔MRI、前列腺特异性膜抗原正电子发射计算机断层扫描(prostate specific membrane antigen positron emission tomography-computed tomography, PSMA PET/CT)或胆碱PET/CT]显示病灶增大[22]。阴性结果定义:(1)病理活检结果为阴性;(2)随访影像检查至少12个月非进展性阴性结果,且至少24个月PSA值未升高[23]

1.2 设备与方法

       采用SIEMENS 3.0 T Skyra及Verio MRI扫描仪,腹部12及8通道相控阵线圈。患者取仰卧位,范围自骶髂关节上缘到盆底,中心定位在耻骨联合上方5 cm。嘱患者扫描前适时饮水,使膀胱呈中度充盈,两台扫描仪均采用规范化的前列腺mpMRI扫描序列及相同标准化的成像协议[13, 24],并定期进行设备校准,扫描序列及参数:横断面和冠状面T2WI序列,TR 4700 ms,TE 65 ms,层厚3 mm,矩阵288×244,FOV 220 mm×220 mm;动态对比增强MRI(dynamic contrast-enhanced MRI, DCE-MRI)序列为T1-VIBE-DYN,TR 3.23 ms,TE 1.38 ms,层厚3 mm,FOV 260 mm×260 mm,静脉注射钆特酸葡胺(中国江苏恒瑞医药股份有限公司),剂量0.1 mmol/kg,使用高压注射器(Spectris Solaris Ep,拜耳医药保健有限公司,德国)以2.5 mL/s的速率注射;DWI扫描为平面回波成像(echoplanar imaging, EPI)序列,TR 2545 ms,TE 72 ms,层厚3 mm,FOV 200 mm×200 mm,b值选择50、800、1400 s/mm2

       通过所采集的两个较低b值DWI进行表观扩散系数(apparent diffusion coefficient, ADC)计算,后根据ADC计算合成高b值DWI[13]。过程:利用b值50、800 s/mm2的DWI计算初始的ADC,再利用ADC值计算DWI(b值=2000 s/mm2)。此过程使用单指数衰减模型(公式1)来描述信号强度S(b)随扩散敏感因子b及ADC的变化[25]。S0是b=0 s/mm2时的信号强度。

1.3 前列腺DWI生成模型的构建

       扩散模型是一类强大的用于学习复杂数据分布概率的生成模型,通过利用正向扩散和反向扩散过程来实现此过程[26]。正向扩散过程将噪声添加到输入数据中,逐渐增加噪声水平,直到数据转化为纯高斯噪声,该过程系统的扰乱数据分布的结构。随后反向扩散过程则从扰乱的数据分布中恢复原始结构,此过程有效地消除正向扩散过程造成的退化效应。最后产生一个高度灵活且易处理的生成模型,可准确地对复杂数据分布进行建模并从随机噪声中还原真实信息。

       去噪扩散概率模型(denoising diffusion probabilistic models, DDMP)是扩散模型在深度学习算法架构下的一种特定形式[27, 28, 29]。本研究采用基于DDMP为网络框架的端到端图像转换方法构建高b值DWI图像生成模型。该模型利用2000例前列腺癌患者的DWI配对数据进行训练,构建高b值(b=2000 s/mm2)DWI生成模型,在此过程中,输出图像以纯高斯噪声作为初始状态,随后通过应用U-Net架构进行迭代优化,将原始扫描获得的b=1400 s/mm2 DWI图像输入后,输出获得b=2000 s/mm2的生成DWI图像,并生成相应ADC图,该架构经过在多种噪声水平下对低分辨率输入图像的训练,且忽略输入图像的采集异质性和质量差异,实现输出图像的高质量提升。

1.4 图像评价及复发风险评分

1.4.1 DWI图像评价

       阅片者在进行评估DWI图像前,为减少评估偏倚,先将计算DWI和生成DWI进行匿名处理,随后缩放裁剪到相同尺度并进行乱序排列,分为两组,确保每组只包含同一患者的一个DWI序列。随后由1名具有13年腹部肿瘤影像诊断经验(副主任医师)及2名分别有5年和7年肿瘤影像诊断经验(主治医师)的放射科医师担任阅片者(Reader1、Reader2及Reader3),在分别间隔两周的前提下对两组图像进行评估。

       阅片者采用Likert量表4分法对图像质量进行评价(1分=差,2分=中等,3分=良好,4分=优秀)[30, 31]。其中:1分,图像质量差(存在严重伪影、图像失真或信号强度差);2分,图像质量一般(有一定伪影、图像中度失真,结构显示欠清);3分,图像质量良好(伪影少,轻度失真,结构较清晰);4分,图像质量优(几乎无伪影,无明显图像失真)。

表1  前列腺病灶局部复发状态及判断方法
Tab. 1  Local recurrence status and determination of prostatic lesions
表2  RT组21例患者有无局部复发情况一般资料比较
Tab. 2  Comparison of general data of 21 patients with and without local recurrence in RT group
表3  RP组42例患者有无局部复发情况一般资料比较
Tab. 3  Comparison of general data of 42 patients with and without local recurrence in RP group
表4  阅片者间DWI图像质量评价比较
Tab. 4  Comparison of DWI image quality evaluation among different reader

1.4.2 PI-RR评分及说明

       每位阅片者在不知晓患者当前检查的临床及病理信息情况下,按照RT和RP后不同的PI-RR系统评估标准对图像进行独立分析。PI-RR评分运用前列腺mpMRI预测局部复发的5分量表[32],对DCE-MRI、DWI和T2WI序列分别进行1~5分评分,以DCE-MRI及DWI为主要评分和升级评分依据,T2WI用于定位可疑病变,提供有价值的解剖信息,不参与最终总评分。PI-RR风险评分1~2分的病灶,表明局部复发的可能性极低和低。RT后患者PI-RR评分3~5分的病灶(图1),可作为挽救治疗前活检的指征。对于RP患者BCR和PSA持续升高,PI-RR为3分,应行核医学检查或对前列腺切除床/膀胱尿道吻合口活检,尤其是PSA>1 ng/mL;PI-RR 评分4~5分的病灶(图2),提示肿瘤复发可能,可无须活检而直接进行挽救性治疗[10, 33]

图1  男,78岁,前列腺癌RT后,原右侧叶瘤床区腺体结构萎缩,左侧叶出现新发肿瘤。1A:T2WI显示前列腺左侧叶不规则稍低信号肿块;1B:DCE-MRI显示病灶呈明显不均匀强化,PI-RR评分4分;1C:计算DWI(经初始ADC图计算获得b=2000 s/mm2的DWI)显示病灶呈稍高信号,边界不清,图像信噪比低;1D:初始ADC(经b值50、800 s/mm2的DWI计算获得)图显示呈稍低信号,根据计算DWI表现PI-RR评分2分;1E:生成DWI(经模型生成b=2000 s/mm2的DWI)显示病灶呈明显高信号,边界显示更清晰;1F:生成ADC(经b值50 s/mm2、生成b=2000 s/mm2 DWI计算获得)图呈明显低信号,根据生成DWI的表现PI-RR评分4分。结合DCE-MRI和生成DWI评分后,PI-RR总评分由初始的4分提升为5分。
图2  男,55岁,前列腺癌RP后膀胱尿道吻合口局部肿瘤复发。2A:T2WI显示尿道吻合口处不均匀增厚;2B:DCE-MRI显示尿道吻合口边缘呈轻度延迟强化,PI-RR评分3分;2C:计算DWI(经初始ADC图计算获得b=2000 s/mm2的DWI)显示病灶呈中高信号,与周围组织界限模糊;2D:初始ADC(经b值50、800 s/mm2的DWI计算获得)图呈低信号,根据计算DWI表现PI-RR评分3分;2E:生成DWI(经模型生成b=2000 s/mm2的DWI)显示病灶呈明显高信号,其边界较清晰;2F:生成ADC(经b值50 s/mm2、生成b=2000 s/mm2 DWI计算获得)图呈明显低信号,根据生成DWI表现PI-RR评分4分。结合DCE-MRI和生成DWI评分后,PI-RR总评分由初始的3分提升为4分。RT:放射治疗;DCE-MRI:动态对比增强MRI;PI-RR:前列腺复发影像报告;DWI:扩散加权成像;ADC:表观扩散系数;RP:根治性前列腺切除术。
Fig. 1  Male, 78 years old, patient was post RT for prostate cancer, glandular structures in the area of the tumor bed in the original right lobe atrophied and a new tumor was seen in the left lobe. 1A: T2WI shows a irregular slightly hypointense lesion in the left lobe of the prostate; 1B: DCE-MRI shows a markedly heterogeneous enhancement of the lesion, PI-RR score of 4; 1C: Calculated DWI (DWI with b=2000 s/mm2 is calculated from the initial ADC map) shows a slightly hyperintense lesion with unclear borders and low image signal-to-noise ratio; 1D: Initial ADC (It's calculated by DWI with b values of 50 s/mm2 and 800s/mm2) map shows a slightly hypointense lesion, based on the calculated DWI performance, PI-RR score of 2; 1E: Generated DWI (DWI with b=2000 s/mm2 is generated by the model) image shows a markedly hyperintense lesion, compared with the calculated DWI image, the boundary is more clearly displayed; 1F: Generated ADC (It's calculated from initial b value of 50 s/mm2 and b=2000 s/mm2) map shows a markedly hypointense lesion, based on the generated DWI performance, PI-RR score of 4. After combining the DCE-MRI and generated DWI scores, the total PI-RR score was raised from 4 to 5.
Fig. 2  Male, 55 years old, patient was post RP for prostate cancer with local tumor recurrence in the vesicourethral anastomosis. 2A: T2WI shows uneven thickening at the urethrovesical anastomosis; 2B: DCE-MRI shows mild delayed enhancement at the edge of the urethrovesical anastomosis, PI-RR score of 3; 2C: Calculated DWI (DWI with b=2000 s/mm2 is calculated from the initial ADC map) shows the lesion with intermediate to high signal intensity, with unclear boundaries with surrounding tissues; 2D: Initial ADC (It's calculated by DWI with b values of 50 s/mm2 and 800 s/mm2) map shows low signal intensity, PI-RR score of 3 based on the calculated DWI performance; 2E: Generated DWI (DWI with b=2000 s/mm2 is generated by the model) image shows the lesion with marked high signal intensity, with clearer boundaries; 2F: Generated ADC (It's calculated from initial b value of 50 s/mm2 and b=2000 s/mm2) map shows marked low signal intensity, PI-RR score of 4 based on the generated DWI performance. After combining the DCE-MRI and generated DWI scores, the total PI-RR score was raised from 3 to 4. RT: radiation therapy; DCE-MRI dynamic contrast-enhanced MRI; PI-RR: Prostate Imaging for Recurrence Reporting; DWI: diffusion weighted imaging; ADC: apparent diffusion coefficient; RP: radical prostatectomy.

1.5 统计学方法

       采用SPSS 23.0及R 4.3.2软件进行数据分析。对所有计量资料进行Kolmogorov-Smirnov检验以确定其分布状态,符合正态分布的资料使用均数±准差(x¯±s)表示,并采用独立样本t检验进行两组间比较;非正态分布的计量资料使用中位数(上下四分位数)[MQ1,Q3)]表示,并采用Mann-Whitney U检验进行两组间比较。计数资料用例表示,两组间比较采用χ2检验。对3位观察者一致性检验采用组内相关系数(intra-class correlation coefficient, ICC)评价,ICC>0.75认为一致性良好。采用多读者多病例受试者工作特征(multi-reader multi-case receiver operating characteristic, MRMC-ROC)曲线评价不同阅片者使用PI-RR评分对前列腺癌RT或RP后局部肿瘤复发预测性能。P<0.05表示差异有统计学意义。

2 结果

2.1 一般资料

       前列腺癌根治性治疗后患者共纳入63例,其中RP组42例(复发11例,无复发31例),RT组21例(复发8例,无复发13例),详见表1

       RT及RP两组发生局部复发的患者BCR时PSA值高于无复发患者(P<0.05),两组的年龄、基线PSA、初诊至BCR时间、国际泌尿病理学会(International Society of Urological Pathology, ISUP)分级、手术后病理T及N分期、外科切缘状态差异无统计学意义(P>0.05),详见表23

2.2 DWI图像质量评价比较

       3位阅片者对RT及RP两组生成DWI组图像质量评价整体优于计算DWI组,差异均具有统计学意义(P<0.05;表4);3位阅片者的生成DWI组图像评价一致性良好,ICC系数为0.871 [95%置信区间(confidence interval, CI):0.838~0.898],计算DWI组评价一致性良好,ICC系数为0.831(95% CI:0.777~0.871)。

2.3 PI-RR总评分比较

       由于PI-RR评分系统对RT与RP后的复发评估方法细则有差别[10],所以分别对RT与RP两组的病例进行分析比较。3位阅片者对RT及RP后生成DWI组与计算DWI组的PI-RR总评分差异有统计学意义(P<0.05)(表56),部分病例根据生成DWI的评分将原PI-RR评分有不同程度的提升(图1、2);3位阅片者对生成DWI组PI-RR总评分一致性较好[ICC系数0.902(95% CI:0.851~0.937)],对计算DWI组PI-RR总评分一致性良好[ICC系数0.878(95% CI:0.81~0.922)]。依据PI-RR总评分检测RT及RP后局部复发的诊断性能,3位阅片者对RT后生成DWI组与计算DWI组MRMC-ROC的曲线下面积(area under the curve, AUC)分别为0.924 vs. 0.792、0.889 vs. 0.793、0.884 vs. 0.783(图3);3位阅片者对RP后生成DWI组与计算DWI组的MRMC-ROC的AUC分别为0.947 vs. 0.893、0.926 vs. 0.853、0.945 vs. 0.843(图4);将RT与RP两组合并后,3位阅片者对生成DWI组与计算DWI组MRMC-ROC的AUC分别为0.930 vs. 0.858、0.912 vs. 0.797、0.928 vs. 0.832(图5)。

图3  RT组患者PI-RR总评分MRMC-ROC曲线图。计算DWI:经初始ADC图计算获得b=2000 s/mm2的DWI;生成DWI:经模型生成b=2000 s/mm2的DWI。RT:放射治疗;PI-RR:前列腺复发影像报告;MRMC-ROC:多读者多病例受试者工作特征;DWI:扩散加权成像;ADC:表观扩散系数。
Fig. 3  MRMC-ROC curve of the total PI-RR score of patients in the RT group. Calculation DWI: DWI with b=2000 s/mm2 was obtained by initial ADC calculation; generated DWI: DWI with b=2000 s/mm2 was generated by the model. MRMC-ROC: multi-reader multi-case receiver operating characteristic; PI-RR: Prostate Imaging for Recurrence Reporting; RT: radiation therapy; DWI: diffusion weighted imaging; ADC: apparent diffusion coefficient.
图4  RP组患者PI-RR总评分MRMC-ROC曲线图。计算DWI:经初始ADC图计算获得b=2000 s/mm2的DWI;生成DWI:经模型生成b=2000 s/mm2的DWI。RP:根治性前列腺切除术;PI-RR:前列腺复发影像报告;MRMC-ROC:多读者多病例受试者工作特征;DWI:扩散加权成像;ADC:表观扩散系数。
Fig. 4  MRMC-ROC curve of the total PI-RR score of patients in the RP group. Calculation DWI: DWI with b=2000 s/mm2 was obtained by initial ADC calculation; generated DWI: DWI with b=2000 s/mm2 was generated by the model. MRMC-ROC: multi-reader multi-case receiver operating characteristic; PI-RR: Prostate Imaging for Recurrence Reporting; RP: radical prostatectomy; DWI: diffusion weighted imaging; ADC: apparent diffusion coefficient.
图5  前列腺癌根治性治疗后总体PI-RR评分MRMC-ROC曲线图。计算DWI:经初始ADC图计算获得b=2000 s/mm2的DWI;生成DWI:经模型生成b=2000 s/mm2的DWI。PI-RR:前列腺复发影像报告;MRMC-ROC:多读者多病例受试者工作特征;DWI:扩散加权成像;ADC:表观扩散系数。
Fig. 5  MRMC-ROC curve of overall PI-RR score after radical prostate cancer treatment. Calculation DWI: DWI with b=2000 s/mm2 was obtained by initial ADC calculation; generated DWI: DWI with b=2000 s/mm2 was generated by the model. MRMC-ROC:Multi-reader multi-case receiver operating characteristic; PI-RR: Prostate Imaging for Recurrence Reporting; DWI: diffusion weighted imaging; ADC: apparent diffusion coefficient.
表5  阅片者间RT组PI-RR总评分比较
Tab. 5  Comparison of total PI-RR scores in the RT group among different reader
表6  阅片者间RP组PI-RR总评分比较
Tab. 6  Comparison of total PI-RR scores in the RP group among different reader

3 讨论

       本研究利用基于扩散模型架构下深度学习算法建立的前列腺高b值DWI图像生成模型,对63例前列腺癌根治性治疗后的高b值DWI图像进行模型输入处理后,发现基于该模型生成的高b值DWI图像质量显著优于传统计算高b值DWI。此外,利用生成DWI进行前列腺癌患者根治性治疗后的PI-RR总评分诊断效能优于传统计算高b值DWI。本研究为改善前列腺癌根治性治疗后高b值DWI图像质量,提升前列腺癌治疗后局部复发诊断效能提供了新思路。

3.1 前列腺高b值DWI图像质量的改进

       提升高b值DWI的图像质量对前列腺疾病定性诊断具有较高的临床价值,国内外学者均对此开展了诸多方面的研究。WERNER等[31]利用复合平均增强图像处理技术将53例前列腺MRI检查中的常规DWI图(b=1000 s/mm2)计算生成获得高b值(b=2000 s/mm2)图像,结果显示该方法显著提高了DWI图像质量,提高临床相关病变的检出率并升级了一部分患者病变的前列腺影像报告和数据系统(Prostate Imaging Reporting and Data System, PI-RADS)评分。在前期研究中,笔者建立了基于生成对抗网络框架下深度学习算法的DWI模型,并利用多中心数据合成前列腺DWI图像,研究结果显示生成的高b值(b=1500 s/mm2)DWI图像质量优于计算DWI图像,同时显著提高了前列腺癌mpMRI的检出率和诊断准确率,得出此类模型具有良好的临床应用价值[20]。本次研究将焦点聚于前列腺癌根治性治疗后MRI随访评估,结果显示采用前列腺癌治疗后的DWI图像,利用生成模型输出的高b值生成DWI图像质量评分的一致性及总体质量评分均优于计算DWI,体现了构建的DWI生成模型具有较好的稳定性和实用性。

3.2 前列腺癌治疗后对DWI图像影响

       前列腺癌根治性RT后可引起前列腺解剖结构的改变,包括腺体体积减小、T2WI弥漫性低信号、高b值DWI显示前列腺信号强度降低,使腺体内良恶性组织区域之间分辨差,增加了医生对病灶的评估难度[34, 35]。但是,在本研究中将RT组的计算DWI图像经过模型处理后,生成的高b值DWI图像能够通过模型去噪和良好的背景抑制效果及减少伪影来显示出肿瘤与非肿瘤的区域,使前列腺区域病灶显示更加清晰。利用这些提升后的图像,可让放射科医生更有依据地对图像进行主观评分[19]。相比于RT治疗的患者其前列腺结构仍然存在,RP后患者前列腺及两侧精囊腺均被切除,原前列腺区域代之为膀胱尿道吻合口[36]。在原始采集DWI信号时前列腺组织的信号呈缺失状态且后方邻近的直肠肠腔结构,以上解剖结构的变化与肠腔磁敏感伪影会形成重叠干扰[37],使RP后获得的计算高b值DWI图像质量比RT后的图像质量要更低,尤其是RP后在膀胱尿道吻合口周围若无局部复发征象,便无法检测相应的病灶,利用模输出的生成DWI图像也为无病灶的图像,尽管如此,生成的DWI图像质量在整体评价结果显示仍优于计算DWI图像,由此可见扩散模型对于周围组织引起的磁敏感伪影干扰具有良好的去除效果。

3.3 生成DWI图像在PI-RR评分应用

       本研究将生成的高b值DWI图像应用到评估前列腺癌患者治疗后有无局部复发的PI-RR评分系统,结果显示3位阅片者使用生成DWI图对两组的PI-RR总评分与计算DWI组的总评分差异均有统计学意义,生成DWI组PI-RR评分对前列腺癌根治性治疗后局部有无复发的预测效能高于计算DWI组。其中RT组有3例患者、RP组4例患者在利用生成的DWI图进行评分后,分别将最终PI-RR总评分提升至4~5分的组别。阅片者1对两组的PI-RR总评分诊断效能稍略高于阅片者2和3,可能因其从事肿瘤相关影像诊断工作时间相对较长,对病灶的识别及诊断经验相对丰富。另外,研究发现无论采用生成DWI或计算DWI进行PI-RR评分时,RT组的PI-RR总评分效能均低于RP组,分析其可能的原因有:患者在接受RT后原前列腺结构显著萎缩,当病灶较小而显示不佳时,阅片医生对病灶甄别难度增大而出现评分判断偏差;也可能因RT组样本数量较少,容易受到偶然误差的影响,导致结果的变异性增加,从而影响诊断效能。尽管PI-RR评分系统对前列腺癌治疗后局部复发的诊断效能较高,但对于前列腺癌根治性治疗后出现BCR患者由于PI-RR评分为1~2分而被排除局部复发的人群,在国外指南中仍然会推荐进行PSMA PET/CT检查来明确是否发生远处转移,以排除局部阴性的表现[38, 39]。对于PI-RR评分3分且伴PSA升高的患者,更加应当积极接受临床的进一步检查,包括有创性检查来明确复发情况[40]

3.4 本研究的局限性

       本研究存在以下不足:(1)由于采用单中心回顾性研究所纳入样本量较小,结果可能会受到一定的选择及偶然偏差影响;(2)对图像质量的分析评价为阅片者的主观判断,缺少定量分析指标;(3)对局部复发判断为阳性的方法主要为后续随访影像及实验室检查,进行病理组织活检例数偏少而可能存在假阳性病例。下一步需要在多中心、大样本量、前瞻性的临床研究中进行不断完善和改进。

4 结论

       综上所述,基于扩散模型生成的前列腺高b值DWI图可应用于对前列腺癌根治性治疗后局部复发的风险评估,为放射科医师在PI-RR系统中的DWI评分增加诊断信心,并有助于临床决策者依据复发风险评分来实现更好的诊疗建议和个体化的治疗。

[1]
冯朝燕, 张配配, 闵祥德. 磁共振成像预测前列腺癌根治术后生化复发的研究进展[J]. 磁共振成像, 2023, 14(5): 186-190, 202. DOI: 10.12015/issn.1674-8034.2023.05.033.
FENG C Y, ZHANG P P, MIN X D. Research progress of magnetic resonance imaging in predicting biochemical recurrence of prostate cancer after radical prostatectomy[J]. Chin J Magn Reson Imag, 2023, 14(5): 186-190, 202. DOI: 10.12015/issn.1674-8034.2023.05.033.
[2]
SCHAEFFER E, SRINIVAS S, ANTONARAKIS E S, et al. NCCN guidelines insights: prostate cancer, version 1.2021[J]. J Natl Compr Canc Netw, 2021, 19(2): 134-143. DOI: 10.6004/jnccn.2021.0008.
[3]
MOTTET N, VAN DEN BERGH R C N, BRIERS E, et al. EAU-EANM-ESTRO-ESUR-SIOG guidelines on prostate cancer-2020 update. part 1: screening, diagnosis, and local treatment with curative intent[J]. Eur Urol, 2021, 79(2): 243-262. DOI: 10.1016/j.eururo.2020.09.042.
[4]
LUKER G D. Imaging biochemical recurrence in prostate cancer[J/OL]. Radiol Imaging Cancer, 2021, 3(4): e219015 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/34328353/. DOI: 10.1148/rycan.2021219015.
[5]
PISANSKY T M, THOMPSON I M, VALICENTI R K, et al. Adjuvant and salvage radiotherapy after prostatectomy: ASTRO/AUA guideline amendment 2018-2019[J]. J Urol, 2019, 202(3): 533-538. DOI: 10.1097/JU.0000000000000295.
[6]
HU C H, QIAO X M, HUANG R P, et al. Development and validation of a multimodality model based on whole-slide imaging and biparametric MRI for predicting postoperative biochemical recurrence in prostate cancer[J/OL]. Radiol Imaging Cancer, 2024, 6(3): e230143 [2024-07-09]. https://pubmed.ncbi.nlm.nih.gov/38758079/. DOI: 10.1148/rycan.230143.
[7]
LUZZAGO S, COLOMBO A, MISTRETTA F A, et al. Multiparametric MRI-based 5-year risk prediction model for biochemical recurrence of prostate cancer after radical prostatectomy[J/OL]. Radiology, 2023, 309(2): e223349 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37987657/. DOI: 10.1148/radiol.223349.
[8]
CICCARESE F, CORCIONI B, BIANCHI L, et al. Clinical application of the new Prostate Imaging for Recurrence Reporting (PI-RR) score proposed to evaluate the local recurrence of prostate cancer after radical prostatectomy[J/OL]. Cancers, 2022, 14(19): 4725 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/36230647/. DOI: 10.3390/cancers14194725.
[9]
HAIDER M A. It's time for a standardized MRI assessment scheme for prostate cancer recurrence[J]. Radiology, 2022, 304(2): 351-352. DOI: 10.1148/radiol.220701.
[10]
PANEBIANCO V, VILLEIRS G, WEINREB J C, et al. Prostate magnetic resonance imaging for local recurrence reporting (PI-RR): international consensus-based guidelines on multiparametric magnetic resonance imaging for prostate cancer recurrence after radiation therapy and radical prostatectomy[J]. Eur Urol Oncol, 2021, 4(6): 868-876. DOI: 10.1016/j.euo.2021.01.003.
[11]
TURKBEY B, ROSENKRANTZ A B, HAIDER M A, et al. Prostate imaging reporting and data system version 2.1: 2019 update of prostate imaging reporting and data system version 2[J]. Eur Urol, 2019, 76(3): 340-351. DOI: 10.1016/j.eururo.2019.02.033.
[12]
TURKBEY B. Better image quality for diffusion-weighted MRI of the prostate using deep learning[J]. Radiology, 2022, 303(2): 382-383. DOI: 10.1148/radiol.212078.
[13]
中华放射学杂志前列腺疾病诊疗工作组, 中华放射学杂志编辑委员会. 前列腺癌MRI检查和诊断共识(第二版)[J]. 中华放射学杂志, 2018, 52(10): 743-750. DOI: 10.3760/cma.j.issn.1005-1201.2018.10.005.
Working Group of Prostate Disease of Chinese Journal of Radiology, Editorial Board of Chinese Journal of Radiology. Consensus on MRI examination and diagnosis of prostate cancer (second edition)[J]. Chin J Radiol, 2018, 52(10): 743-750. DOI: 10.3760/cma.j.issn.1005-1201.2018.10.005.
[14]
BISCHOFF L M, PEETERS J M, WEINHOLD L, et al. Deep learning super-resolution reconstruction for fast and motion-robust T2-weighted prostate MRI[J/OL]. Radiology, 2023, 308(3): e230427 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37750774/. DOI: 10.1148/radiol.230427.
[15]
PARK J C, PARK K J, PARK M Y, et al. Fast T2-weighted imaging with deep learning-based reconstruction: evaluation of image quality and diagnostic performance in patients undergoing radical prostatectomy[J]. J Magn Reson Imaging, 2022, 55(6): 1735-1744. DOI: 10.1002/jmri.27992.
[16]
TAJIMA T, AKAI H, SUGAWARA H, et al. Feasibility of accelerated whole-body diffusion-weighted imaging using a deep learning-based noise-reduction technique in patients with prostate cancer[J]. Magn Reson Imaging, 2022, 92: 169-179. DOI: 10.1016/j.mri.2022.06.014.
[17]
JEONG J, YEOM S K, CHOI I Y, et al. Deep learning image reconstruction of diffusion-weighted imaging in evaluation of prostate cancer focusing on its clinical implications[J]. Quant Imaging Med Surg, 2024, 14(5): 3432-3446. DOI: 10.21037/qims-23-1379.
[18]
LIN Y, YILMAZ E C, BELUE M J, et al. Evaluation of a cascaded deep learning-based algorithm for prostate lesion detection at biparametric MRI[J/OL]. Radiology, 2024, 311(2): e230750 [2024-07-09]. https://pubmed.ncbi.nlm.nih.gov/38713024/. DOI: 10.1148/radiol.230750.
[19]
UEDA T, OHNO Y, YAMAMOTO K, et al. Deep learning reconstruction of diffusion-weighted MRI improves image quality for prostatic imaging[J]. Radiology, 2022, 303(2): 373-381. DOI: 10.1148/radiol.204097.
[20]
HU L, ZHOU D W, ZHA Y F, et al. Synthesizing high-b-value diffusion-weighted imaging of the prostate using generative adversarial networks[J/OL]. Radiol Artif Intell, 2021, 3(5): e200237 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/34617025/. DOI: 10.1148/ryai.2021200237.
[21]
EPSTEIN J I, EGEVAD L, AMIN M B, et al. The 2014 international society of urological pathology (ISUP) consensus conference on gleason grading of prostatic carcinoma: definition of grading patterns and proposal for a new grading system[J]. Am J Surg Pathol, 2016, 40(2): 244-252. DOI: 10.1097/PAS.0000000000000530.
[22]
PRIVÉ B M, ISRAËL B, JANSSEN M J R, et al. Multiparametric MRI and 18F-PSMA-1007 PET/CT for the detection of clinically significant prostate cancer[J/OL]. Radiology, 2024, 311(2): e231879 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/38771185/. DOI: 10.1148/radiol.231879.
[23]
PECORARO M, TURKBEY B, PURYSKO A S, et al. Diagnostic accuracy and observer agreement of the MRI Prostate Imaging for Recurrence Reporting assessment score[J]. Radiology, 2022, 304(2): 342-350. DOI: 10.1148/radiol.212252.
[24]
ISRAËL B, LEEST M V, SEDELAAR M, et al. Multiparametric magnetic resonance imaging for the detection of clinically significant prostate cancer: what urologists need to know. part 2: interpretation[J]. Eur Urol, 2020, 77(4): 469-480. DOI: 10.1016/j.eururo.2019.10.024.
[25]
MAAS M C, FÜTTERER J J, SCHEENEN T W J. Quantitative evaluation of computed high B value diffusion-weighted magnetic resonance imaging of the prostate[J]. Invest Radiol, 2013, 48(11): 779-786. DOI: 10.1097/RLI.0b013e31829705bb.
[26]
KAZEROUNI A, AGHDAM E K, HEIDARI M, et al. Diffusion models in medical imaging: a comprehensive survey[J/OL]. Med Image Anal, 2023, 88: 102846 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37295311/. DOI: 10.1016/j.media.2023.102846.
[27]
KHADER F, MÜLLER-FRANZES G, TAYEBI ARASTEH S, et al. Denoising diffusion probabilistic models for 3D medical image generation[J/OL]. Sci Rep, 2023, 13(1): 7303 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37147413/. DOI: 10.1038/s41598-023-34341-2.
[28]
ESCHWEILER D, YILMAZ R, BAUMANN M, et al. Denoising diffusion probabilistic models for generation of realistic fully-annotated microscopy image datasets[J/OL]. PLoS Comput Biol, 2024, 20(2): e1011890 [2024-07-09]. https://pubmed.ncbi.nlm.nih.gov/38377165/. DOI: 10.1371/journal.pcbi.1011890.
[29]
GUO Z Y, LIU J, WANG Y L, et al. Diffusion models in bioinformatics and computational biology[J]. Nat Rev Bioeng, 2024, 2(2): 136-154. DOI: 10.1038/s44222-023-00114-9.
[30]
HU L, ZHOU D W, FU C X, et al. Calculation of apparent diffusion coefficients in prostate cancer using deep learning algorithms: a pilot study[J/OL]. Front Oncol, 2021, 11: 697721 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/34568027/. DOI: 10.3389/fonc.2021.697721.
[31]
WERNER S, ZINSSER D, ESSER M, et al. Enhanced image processing using complex averaging in diffusion-weighted imaging of the prostate: the impact on image quality and lesion detectability[J/OL]. Diagnostics, 2023, 13(14): 2325 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37510071/. DOI: 10.3390/diagnostics13142325.
[32]
ABREU-GOMEZ J, DIAS A B, GHAI S. PI-RR: the Prostate Imaging for Recurrence Reporting system for MRI assessment of local prostate cancer recurrence after radiation therapy or radical prostatectomy-a review[J]. AJR Am J Roentgenol, 2023, 220(6): 852-861. DOI: 10.2214/AJR.22.28665.
[33]
LOWRANCE W T, BREAU R H, CHOU R, et al. Advanced prostate cancer: AUA/ASTRO/SUO guideline PART I[J]. J Urol, 2021, 205(1): 14-21. DOI: 10.1097/JU.0000000000001375.
[34]
PATEL P, MATHEW M S, TRILISKY I, et al. Multiparametric MR imaging of the prostate after treatment of prostate cancer[J]. Radiographics, 2018, 38(2): 437-449. DOI: 10.1148/rg.2018170147.
[35]
GAUR S, TURKBEY B. Prostate MR imaging for posttreatment evaluation and recurrence[J]. Radiol Clin North Am, 2018, 56(2): 263-275. DOI: 10.1016/j.rcl.2017.10.008.
[36]
朱学华, 邵立智, 颜野, 等. 前列腺癌根治术后生化复发相关MRI影像组学特征的研究[J]. 中国微创外科杂志, 2023, 23(5): 359-366. DOI: 10.3969/j.issn.1009-6604.2023.05.008.
ZHU X H, SHAO L Z, YAN Y, et al. MRI-based radiomics features associated with biochemical recurrence of prostate cancer after radical prostatectomy[J]. Chin J Minim Invasive Surg, 2023, 23(5): 359-366. DOI: 10.3969/j.issn.1009-6604.2023.05.008.
[37]
SATHIADOSS P, HAROON M, OSMAN H, et al. Comparison of 5 rectal preparation strategies for prostate MRI and impact on image quality[J]. J L'association Can Des Radiol, 2022, 73(2): 346-354. DOI: 10.1177/08465371211033753.
[38]
AWIWI M O, GJONI M, VIKRAM R, et al. MRI and PSMA PET/CT of biochemical recurrence of prostate cancer[J/OL]. Radiographics, 2023, 43(12): e230112 [2024-04-20]. https://pubmed.ncbi.nlm.nih.gov/37999983/. DOI: 10.1148/rg.230112.
[39]
DE VISSCHERE P J L, STANDAERT C, FÜTTERER J J, et al. A systematic review on the role of imaging in early recurrent prostate cancer[J]. Eur Urol Oncol, 2019, 2(1): 47-76. DOI: 10.1016/j.euo.2018.09.010.
[40]
VAN DEN BROECK T, VAN DEN BERGH R C N, ARFI N, et al. Prognostic value of biochemical recurrence following treatment with curative intent for prostate cancer: a systematic review[J]. Eur Urol, 2019, 75(6): 967-987. DOI: 10.1016/j.eururo.2018.10.011.

上一篇 IVIM全容积定量参数与直肠腺癌神经脉管侵犯、MSI状态及Ki-67指数的相关性研究
下一篇 DCE-MRI结合DWI对前列腺癌Ki-67表达和Gleason评分的预测价值
  
诚聘英才 | 广告合作 | 免责声明 | 版权声明
联系电话:010-67113815
京ICP备19028836号-2