切换至 "中华医学电子期刊资源库"

中华结直肠疾病电子杂志 ›› 2026, Vol. 15 ›› Issue (03) : 232 -243. doi: 10.3877/cma.j.issn.2095-3224.2026.03.006

论著

基于磁共振成像的深度学习方法在直肠癌术前TN分期中的应用研究
黄子强1, 马晓璐1, 沈浮1, 王颢2, 邵成伟1, 张卫2, 陆建平1, 陆海迪1,()   
  1. 1 200433 海军军医大学第一附属医院(上海长海医院)影像医学科
    2 200433 海军军医大学第一附属医院(上海长海医院)影像肛肠外科
  • 收稿日期:2026-01-24 出版日期:2026-06-25
  • 通信作者: 陆海迪
  • 基金资助:
    上海申康医院发展中心第二轮《促进市级医院临床技能与临床创新三年行动计划》研究型医师创新转化能力培训项目(SHDC2023CRD023)

Application of MRI-based deep learning method in preoperative TN staging of rectal cancer

Ziqiang Huang1, Xiaolu Ma1, Fu Shen1, Hao Wang2, Chengwei Shao1, Wei Zhang2, Jianping Lu1, Haidi Lu1,()   

  1. 1 Department of Radiology, Shanghai 200433, China
    2 Department of Colorectal Surgery, the First Affiliated Hospital of Naval Medical University (Changhai Hospital), Shanghai 200433, China
  • Received:2026-01-24 Published:2026-06-25
  • Corresponding author: Haidi Lu
引用本文:

黄子强, 马晓璐, 沈浮, 王颢, 邵成伟, 张卫, 陆建平, 陆海迪. 基于磁共振成像的深度学习方法在直肠癌术前TN分期中的应用研究[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(03): 232-243.

Ziqiang Huang, Xiaolu Ma, Fu Shen, Hao Wang, Chengwei Shao, Wei Zhang, Jianping Lu, Haidi Lu. Application of MRI-based deep learning method in preoperative TN staging of rectal cancer[J/OL]. Chinese Journal of Colorectal Diseases(Electronic Edition), 2026, 15(03): 232-243.

目的

探索基于MRI的深度学习(DL)分割以及分类模型在评估直肠癌术前TN分期方面的应用价值。

方法

回顾性分析2019年1月至2023年12月海军军医大学第一附属医院术后病理证实为直肠癌的患者临床、病理及治疗前MRI资料。根据时间顺序依次分为训练集和测试集。对纳入研究对象的直肠高分辨T2WI首先采用深度学习重建图像(DLR)技术进行重建优化。在获取的DLR的T2WI图像上分别采用最小靶区勾画(Method 1)和最大靶区勾画(Method 2)两种方法手动逐层勾画病灶感兴趣的区域,构建并比较DL分割模型。随后基于最佳分割方法结果构建直肠癌术前TN分期的DL分类模型,并对其进行综合评估。

结果

本研究纳入326例患者,以2019年—2020年143例为训练集,2021年—2023年183例为测试集。测试集显示,Method 2的DL分割模型戴斯相似系数(DSC)、95%豪斯多夫距离(HD95)、平均表面距离(ASD)中位数分别为0.796、7.730 mm和1.420 mm,优于Method 1(0.708、10.931 mm和2.388 mm),且DSC更高(P<0.001)、ASD更小(P=0.013)。采用Method 2的分割结果构建DL分类模型。T分期任务中177例测试结果显示,模型AUC(95%CI)为0.957(0.932~0.975),显著高于主观评价的0.811(0.733~0.882),其诊断特异度、敏感度、准确度依次为91.5%、95.3%、93.8%;N分期任务中182例测试结果显示,AUC(95%CI)为0.837(0.768~0.897),显著优于主观评价的0.625(0.558~0.803),其诊断特异度、敏感度、准确度分别达98.6%、63.9%、91.8%;Delong检验均P<0.001,且该模型拟合良好,临床净收益更高。

结论

DLR优化的MRI图像结合最大靶区勾画的DL分割模型,可以精准发掘定量信息,显著提升直肠癌术前TN分期的诊断效能。

Objective

To explore the application value of deep learning (DL)-based segmentation and classification models using MRI for preoperative TN staging of rectal cancer.

Methods

A retrospective analysis was conducted on clinical, pathological, and pre-treatment MRI data from patients with postoperatively pathologically confirmed rectal cancer at the First Affiliated Hospital of Naval Medical University between January 2019 and December 2023. Patients were chronologically divided into training and testing cohorts. High-resolution rectal T2-weighted imaging (T2WI) for all enrolled subjects was first reconstructed and optimized using Deep Learning Reconstruction (DLR) technology. On the DLR-reconstructed T2WI images, the lesion region of interest was manually delineated slice-by-slice using two methods: minimal target delineation (Method 1) and maximal target delineation (Method 2). DL segmentation models were constructed and compared. Subsequently, a DL classification model for preoperative TN staging of rectal cancer was built based on the optimal segmentation method and comprehensively evaluated.

Results

This study included 326 patients, with 143 patients from 2019~2020 as the training set and 183 patients from 2021~2023 as the testing set. The test set showed that the DL segmentation model of Method 2 achieved median DSC, HD95, and ASD of 0.796, 7.730 mm, and 1.420 mm, respectively, which were superior to Method 1 (0.708, 10.931 mm, and 2.388 mm), with higher DSC (P<0.001) and smaller ASD (P=0.013). The DL classification model was constructed using the segmentation results of Method 2. In the T-staging task, 177 test cases showed that the model achieved an AUC (95%CI) of 0.957 (0.932~0.975), significantly higher than subjective evaluation at 0.811 (0.733~0.882). The diagnostic specificity, sensitivity and accuracy of the model were 91.5%, 95.3% and 93.8%, respectively. In the N-staging task, 182 test cases showed an AUC (95%CI) of 0.837 (0.768~0.897), significantly better than subjective evaluation at 0.625 (0.558~0.803), with diagnostic specificity, sensitivity and accuracy of 98.6%, 63.9% and 91.8%, respectively. DeLong tests showed P<0.001 for both, and the model demonstrated good calibration and higher clinical net benefit.

Conclusion

DLR-optimized MRI images combined with the DL segmentation model using maximal target delineation can accurately extract quantitative information and significantly improve the diagnostic efficacy of preoperative TN staging of rectal cancer.

图1 DLR图像及靶区勾画方法示例。1A、1B:T2WI斜横断面序列图像,1A为原始高分辨T2WI图像,1B为DLR重建图像。1C、1D:两种靶区勾画方法示意图,1C为最小轮廓法(沿病灶边缘最明确的实性边界勾画,不含边缘模糊区域),1D为最大轮廓法(沿病灶最大范围勾画,包括肠壁外围的模糊软组织影)
表1 深度学习分割算法的参数配置
表2 深度学习分类算法的参数配置
表3 训练集及测试集一般资料比较[例(%),MQ1Q3)]
表4 测试集两种不同分割方法的DL分割模型结果
图2 直肠癌术前TN分期的ROC曲线图。2A:直肠癌术前T分期;2B:直肠癌术前N分期,蓝色和橙色曲线分别代表DL分类模型和主观评价
表5 测试集DL分类模型与主观评价的ROC曲线分析
图3 直肠癌术前T分期及N分期诊断效能比较。3A:基于深度学习模型;3B:基于主观评价
图4 深度学习与主观评价TN分期的DCA。4A:T分期;4B:N分期
图5 深度学习模型矫正曲线。5A:T分期;5B:N分期
图6 深度学习分类模型类激活图可视化分析。6A:直肠癌T分期;6B:直肠癌N分期。T3-4及N1-2为阳性预测病例,T1-2及N0为阴性预测病例,从左至右依次为原始图像、融合图像及类激活图。类激活图显示,阳性预测的激活区域较阴性预测更为显著
[1]
Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024[J]. CA Cancer J Clin, 2024, 74(1): 12-49.
[2]
闫超, 陕飞, 李子禹. 2020年中国与全球结直肠癌流行概况分析[J]. 中华肿瘤杂志, 2023, 45(3): 221-229.
[3]
Mi M, Weng S, Xu Z, et al. CSCO guidelines for colorectal cancer version 2023: updates and insights[J]. Chin J Cancer Res, 2023, 35(3): 233-238.
[4]
常莉娜, 刘囡囡. DWI联合DCE-MRI定量参数对局部进展期直肠癌新辅助治疗后病理完全缓解的预测价值[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(2): 125-129.
[5]
Nougaret S, Reinhold C, Mikhael HW, et al. The use of MR imaging in treatment planning for patients with rectal carcinoma: have you checked the“DISTANCE”?[J]. Radiology, 2013, 268(2): 330-344.
[6]
Wetzel A, Viswanath S, Gorgun E, et al. Staging and restaging of rectal cancer with MRI: a pictorial review[J]. Semin Ultrasound CT MR, 2022, 43(6):441-454.
[7]
Horvat N, Bates D, Petkovska I. Novel imaging techniques of rectal cancer: what do radiomics and radiogenomics have to offer? A literature review[J]. Abdom Radiol (NY), 2019, 44(11): 3764-3774.
[8]
Xia W, Li D, He W, et al. Multicenter evaluation of a weakly supervised deep learning model for lymph node diagnosis in rectal cancer at MRI[J]. Radiol Artif Intell, 2024, 6(2): e230152.
[9]
石晟铭, 肖玲清, 马佳琪, 等. 直肠癌基于磁共振成像的深度学习研究进展[J]. 磁共振成像, 2024, 15(3): 218-222.
[10]
陆海迪. 基于MRI的人工智能方法在直肠癌术前评估中应用研究[D]. 上海: 中国人民解放军海军军医大学, 2025.
[11]
Hou M, Zhou L, Sun J. Deep-learning-based 3D super-resolution MRI radiomics model: superior predictive performance in preoperative T-staging of rectal cancer[J]. Eur Radiol, 2023, 33(1): 1-10.
[12]
吴树剑, 俞咏梅, 范莉芳, 等. 基于深度学习的影像组学预测直肠癌T2与T3分期[J]. 磁共振成像, 2023, 14(11): 84-89, 102.
[13]
Wei Y, Wang H, Chen Z, et al. Deep learning-based multiparametric MRI model for preoperative T-stage in rectal cancer[J]. J Magn Reson Imaging, 2024, 59(3): 1083-1092.
[14]
Radici L, Ferrario S, Borca VC, et al. Implementation of a commercial deep learning-based auto segmentation software in radiotherapy: evaluation of effectiveness and impact on workflow[J]. Life (Basel), 2022, 12(12): 2088.
[15]
唐瑗玲, 王辛. 图像分割在直肠癌放疗领域的应用及进展[J]. 中华放射肿瘤学杂志, 2024, 33(9): 859-863.
[16]
Zhang K, Yang X, Cui Y, et al. Imaging segmentation mechanism for rectal tumors using improved U-Net[J]. BMC Med Imaging, 2024, 24(1): 95.
[17]
Pan X, Tong K, Yan C, et al. Research progress on colorectal cancer identification based on convolutional neural network[J]. J Biomedical Eng, 2024, 41(4): 854-860.
[18]
周振, 沈浮, 陆海迪, 等. 基于高分辨T2WI的影像组学中不同分割方法对直肠癌术前T分期的影响[J]. 中国医学计算机成像杂志, 2021, 27(3): 225-230.
[19]
胡振远, 魏炜, 胡文鐘, 等. 基于磁共振多参数图像融合及DenseNet网络的脑胶质瘤IDH突变预测研究[J]. 磁共振成像, 2023, 14(7): 10-17.
[20]
帕哈提·吐逊江, 赵伟, 罕迦尔别克·库锟, 等. 基于扩散加权成像的密集卷积神经网络模型预测急性缺血性脑卒中TOAST病因分型的研究[J]. 中华放射学杂志, 2024, 58(10): 1015-1020.
[21]
Benson AB, Venook AP, Al-Hawary MM, et al. Rectal cancer, version 2.2022, NCCN clinical practice guidelines in oncology[J]. J Natl Compr Canc Netw, 2022, 20(10): 1139-1167.
[22]
Yan H, Yang H, Jiang P, et al. A radiomics model based on T2WI and clinical indexes for prediction of lateral lymph node metastasis in rectal cancer[J]. Asian J Surg, 2024, 47(1): 450-458.
[23]
Borgheresi A, De Muzio F, Agostini A, et al. Lymph nodes evaluation in rectal cancer: where do we stand and future perspective[J]. J Clin Med, 2022, 11(9).
[24]
欧阳淦露, 王辛. 影像组学和人工智能在直肠癌疗效预测和评估中的作用[J]. 中华放射肿瘤学杂志, 2023, 32(4): 360-364.
[25]
Jiang X, Hu Z, Wang S, et al. Deep learning for medical image-based cancer diagnosis[J]. Cancers (Basel), 2023, 15(14).
[26]
李清扬, 张晓燕, 孙应实. 影像组学与深度学习在评价直肠癌淋巴结状态中的应用[J]. 中华放射学杂志, 2023, 57(2): 231-234.
[27]
Soffer S, Ben-Cohen A, Shimon O, et al. Convolutional neural networks for radiologic images: a radiologist’s guide[J]. Radiology, 2019, 290(3): 590-606.
[28]
严福华. 深度学习MRI重建算法的临床应用和发展前景[J]. 磁共振成像, 2023, 14(5): 8-10.
[29]
Peng W, Wan L, Tong X, et al. Prospective and multi-reader evaluation of deep learning reconstruction-based accelerated rectal MRI: image quality, diagnostic performance, and reading time[J]. Eur Radiol, 2024, 34(11): 7438-7449.
[30]
Wang S, Cao G, Wang Y, et al. Review and prospect: artificial intelligence in advanced medical imaging[J]. Front Radiol, 2021, 1: 781868.
[31]
Tang H, Hong M, Yu L, et al. Deep learning reconstruction for lumbar spine MRI acceleration: a prospective study[J]. Eur Radiol Exp, 2024, 8(1): 67.
[32]
DeSilvio T, Antunes JT, Bera K, et al. Region-specific deep learning models for accurate segmentation of rectal structures on post-chemoradiation T2w MRI: a multi-institutional, multi-reader study[J]. Front Med (Lausanne), 2023, 10: 1149056.
[33]
Zhang X, Zhong L, Zhang B, et al. The effects of volume of interest delineation on MRI-based radiomics analysis: evaluation with two disease groups[J]. Cancer Imaging, 2019, 19(1): 89.
[34]
Kocak B, Ates E, Durmaz ES, et al. Influence of segmentation margin on machine learning-based high-dimensional quantitative CT texture analysis: a reproducibility study on renal clear cell carcinomas[J]. Eur Radiol, 2019, 29(9): 4765-4775.
[35]
Bogveradze N, Snaebjornsson P, Grotenhuis BA, et al. MRI anatomy of the rectum: key concepts important for rectal cancer staging and treatment planning[J]. Insights Imaging, 2023, 14(1): 13.
[36]
Yuan Y, Lu H, Ma X, et al. Is rectal filling optimal for MRI-based radiomics in preoperative T staging of rectal cancer?[J]. Abdom Radiol (NY), 2022, 47(5): 1741-1749.
[37]
朱钰, 欧阳治强, 单海燕, 等. 基于MRI的人工智能在直肠癌中的应用进展[J]. 磁共振成像, 2023, 14(9): 176-180.
[38]
Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation[C]//: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, 2015. Springer.
[39]
Li D, Chu X, Cui Y, et al. Improved U-Net based on contour prediction for efficient segmentation of rectal cancer[J]. Comput Methods Programs Biomed, 2022, 213: 106493.
[40]
Alani M, Jalab H A, Pars S, et al. Enhanced U-Net-Based deep learning model for automated segmentation of organoid images[J]. Bioengineering (Basel), 2025, 12(11).
[41]
吴亚平, 方婷, 魏焕焕, 等. 级联VB-Net分割模型用于急性缺血性脑卒中患者扩散加权成像中缺血核心分割的研究[J]. 中华放射学杂志, 2022, 56(1): 25-29.
[42]
Hua R, Huo Q, Gao Y, et al. Segmenting brain tumor using cascaded V-Nets in multimodal MR images[J]. Front Comput Neurosci, 2020, 14: 9.
[43]
Conti G, Calderan L, Quintero SL, et al. Tumor and peritumoral adipose tissue crosstalk: de-differentiated adipocytes influence spread of colon carcinoma cells[J]. Tissue Cell, 2023, 80: 101990.
[44]
Yang M, Yang M, Yang L, et al. Deep learning for MRI lesion segmentation in rectal cancer[J]. Front Med (Lausanne), 2024, 11: 1394262.
[45]
Men K, Boimel P, Janopaul-Naylor J, et al. Cascaded atrous convolution and spatial pyramid pooling for more accurate tumor target segmentation for rectal cancer radiotherapy[J]. Phys Med Biol, 2018, 63(18): 185016.
[46]
Li J, Zhou Y, Wang P, et al. Deep transfer learning based on magnetic resonance imaging can improve the diagnosis of lymph node metastasis in patients with rectal cancer[J]. Quant Imaging Med Surg, 2021, 11(6): 2477-2485.
[47]
Wan L, Hu J, Chen S, et al. Prediction of lymph node metastasis in stage T1-2 rectal cancers with MRI-based deep learning[J]. Eur Radiol, 2023, 33(5): 3638-3646.
[1] 佘欣怡, 万玮祺, 刘广健, 黄玉笑, 胡金枝, 覃斯. 超声造影定量评估结直肠癌化疗后肝窦阻塞综合征的前瞻性临床研究[J/OL]. 中华医学超声杂志(电子版), 2026, 23(02): 151-161.
[2] 程妹, 金亚彬, 马丁瑞, 程昊, 詹玉莲, 周丹. 基于深度学习的MRI图像分析在乳腺癌诊疗中的应用[J/OL]. 中华乳腺病杂志(电子版), 2026, 20(03): 169-173.
[3] 鲁峰刚, 刘宇, 张毅, 王秦玉. 新辅助放化疗直肠癌根治术患者术后吻合口漏影响因素分析及风险预测模型构建[J/OL]. 中华普外科手术学杂志(电子版), 2026, 20(04): 347-351.
[4] 赵军抗, 张前进, 庄惠杰. 腹腔镜直肠癌根治术保留左结肠动脉的疗效及对预后的影响[J/OL]. 中华普外科手术学杂志(电子版), 2026, 20(03): 257-260.
[5] 叶子兴, 李红运, 李英杰, 王娟, 孙玉姣, 王惠珍, 贺蕾, 吕建阳, 张洪波, 张彤彤, 陈雯雯, 杨晓宇, 申川, 纪志刚. 深度学习模型在膀胱镜下膀胱肿瘤分割中的应用[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2026, 20(04): 399-406.
[6] 黄楚曦, 吴卓. 人工智能在膀胱癌影像中的应用进展[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2026, 20(03): 241-247.
[7] 王浩, 辛才进, 崔静, 傅鹤鑫, 朱长仁, 汪刘华. 错配修复蛋白在早发性结直肠癌中的表达及意义[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(03): 209-214.
[8] 孟春孜, 章群, 钱晓萍. 晚期结直肠癌三线免疫联合方案与单药化疗或单药靶向治疗的疗效对比Meta分析[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(03): 215-223.
[9] 贾宇飞, 王哲学, 王飞, 蔡昂书, 姚占胜. 中性粒细胞与淋巴细胞比值(NLR)和结直肠癌幸存者全因死亡的关联研究:基于NHANES 1999—2018数据的分析[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(03): 224-231.
[10] 马洪庆, 周超熙, 赵泽铭, 胡旭华, 张建锋, 王贵英. 完全腹腔镜下双荷包吻合技术在直肠癌根治术中的应用(附视频)[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(02): 189-192.
[11] 张婧萱, 郭佳星, 李紫红, 陈怡颖, 乔媛, 宝莹娜. 多线治疗失败后31例复发转移性结直肠癌局部挽救性放射治疗的研究[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(03): 206-211.
[12] 董妮, 肖康, 王齐成. 经直肠超声多参数联合检测对中低位直肠癌淋巴结转移的术前预测价值[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(03): 212-215.
[13] 梁玮龙, 王鹏飞, 高怀新. 氟比洛芬酯联合腹横肌平面阻滞在腹腔镜结直肠癌根治术中的麻醉效果及对COX-2、MMP-9水平的影响[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(03): 256-261.
[14] 陈小坤, 杜顺达. 影像组学在肝细胞癌中的应用进展及挑战[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(02): 97-100.
[15] 徐林, 简讯, 刘道权, 易东, 鄢华. 人工智能在心脑血管内介入诊疗中的应用进展——从单一影像模式到多模态影像融合[J/OL]. 中华脑血管病杂志(电子版), 2026, 20(03): 334-341.
阅读次数
全文


摘要


AI


AI小编
你好!我是《中华医学电子期刊资源库》AI小编,有什么可以帮您的吗?