MRI上的尿道轮廓:多学科共识教育图谱与人工智能基准测试的参考标准
本研究由四名前列腺MRI亚专科专家对来自6个中心的71例患者MRI数据生成尿道共识轮廓,建立参考标准,并开发深度学习尿道分割模型;在4例PURE-MRI病例中,AI模型中位Dice为0.41(医师为0.33)、覆盖率为81%(医师为36%)、最大2D HD为1.8 mm(医师为1.6 mm),在全参考数据集中AI表现为Dice 0.40、覆盖率89%、最大2D HD 2.0 mm。
要点
- 前列腺癌放疗中尿道是推荐避开的结构,但亚专科医师在影像上准确识别尿道仍有困难。
- AI尿道分割模型在4例PURE-MRI病例中中位Dice 0.41、覆盖率81%、最大2D HD 1.8 mm。
- 全参考数据集71例中AI表现为Dice 0.40、覆盖率89%、最大2D HD 2.0 mm。
核心结论
- 建立了尿道分割的多学科共识基准,深度学习模型表现与专科医师相当,并在多机构间显示一致结果;其作为临床决策支持工具在前列腺癌放疗计划和剂量-毒性关联研究中显示出前景。
研究设计与样本
- 多学科共识参考标准数据集:4名前列腺MRI亚专科专家对来自6个中心的71例患者MRI数据生成尿道共识轮廓;其中4例曾用于国际研究PURE-MRI(62名医师勾画前列腺和尿道);另用独立训练数据集(151例患者,1个中心)开发深度学习尿道分割AI模型。
诊断标准与关键数值
- 以多学科专家共识尿道轮廓为参考标准,采用Dice、尿道覆盖率百分比和最大2D(轴位、平面内)Hausdorff距离(HD)评估;在4例PURE-MRI病例中,AI模型中位Dice 0.41(医师0.33)、覆盖率81%(医师36%)、最大2D HD 1.8 mm(医师1.6 mm);在全参考数据集中,AI Dice 0.40、覆盖率89%、最大2D HD 2.0 mm。
临床应用与阅片要点
- 作者认为该深度学习模型作为临床决策支持工具,在前列腺癌放疗计划中实现准确可靠的尿道分割以及剂量-毒性关联研究中显示出前景。
局限性
- 原文在PURE-MRI 4例病例中报告AI最大2D HD为1.8 mm,而医师为1.6 mm。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
INTRODUCTION: The urethra is a recommended avoidance structure for prostate cancer treatment. However, even subspecialist physicians often struggle to accurately identify it on available imaging. Automated segmentation tools show promise, but a lack of reliable ground truth or appropriate evaluation standards has hindered validation and clinical adoption. This study aims to establish a reference-standard dataset with expert consensus contours, define clinically meaningful evaluation metrics, and assess the performance and generalizability of a deep-learning-based segmentation model.
MATERIALS AND METHODS: A multidisciplinary panel of four experienced subspecialists in prostate MRI generated consensus urethra contours on MRI data for 71 patients from 6 centers, establishing a reference standard. Four of these patients were previously used in an international study (PURE-MRI) where 62 physicians contoured the prostate and urethra. Using an independent training dataset (n = 151 patients, 1 center), we developed a deep-learning AI model for urethra segmentation. We evaluated the AI tool in the consensus reference dataset and compared it to human performance using Dice, percent urethra coverage, and maximum 2D (axial, in-plane) Hausdorff Distance (HD) from the reference standard.
RESULTS: The AI model outperformed most physicians, achieving median Dice of 0.41 (vs. 0.33 for physicians), Coverage of 81 % (vs. 36 %), and Max 2D HD of 1.8 mm (vs. 1.6 mm) in the four PURE-MRI cases. In the full reference dataset, AI performance remained consistent, with Dice of 0.40, Coverage of 89 %, and Max 2D HD of 2.0 mm, indicating strong generalizability across a broader patient population and more varied imaging conditions.
CONCLUSION: We established a multidisciplinary consensus benchmark for segmentation of the urethra. The deep-learning model performs comparably to specialist physicians and demonstrates consistent results across multiple institutions. It shows promise as a clinical decision-support tool for accurate and reliable urethra segmentation in prostate cancer radiotherapy planning and studies of dose-toxicity associations.
引用
- Song Y, Nguyen L, Dornisch AM, et al. Urethra contours on MRI: Multidisciplinary consensus educational atlas and reference standard for artificial intelligence benchmarking. Radiother Oncol. 2025. PMID: 41429726. doi:10.1016/j.radonc.2025.111231
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。