卷积神经网络在预测肾透明细胞癌全视野数字切片肿瘤坏死中的应用研究

doi:10.3877/cma.j.issn.1674-3253.2026.04.004

中华腔镜泌尿外科杂志(电子版) ›› 2026, Vol. 20 ›› Issue (04) : 389 -398. doi: 10.3877/cma.j.issn.1674-3253.2026.04.004

临床研究

卷积神经网络在预测肾透明细胞癌全视野数字切片肿瘤坏死中的应用研究

李佳颖¹, 刘溪², 陈宇航³, 桂程鹏⁴, 习明², 陈炜¹, 罗俊航¹^,^†(

)

¹510080　广州，中山大学附属第一医院泌尿外科
²510000　广东，广州市花都区人民医院泌尿外科
³300060　天津市肿瘤医院泌尿外科
⁴510030　广东，广州医科大学附属第一医院泌尿外科

收稿日期:2025-08-29 出版日期:2026-08-01
通信作者: 罗俊航
基金资助:
国家自然科学基金面上项目(82573729，82373433); 广州市科技计划项目重点研发计划(2025B03J0091)

Application of convolutional neural networks in predicting tumor necrosis from whole-slide images of clear cell renal cell carcinoma

Jiaying Li¹, Xi Liu², Yuhang Chen³, Chengpeng Gui⁴, Ming Xi², Wei Chen¹, Junhang Luo¹^,^†()

¹Department of Urology, the First Affiliated Hospital of Sun Yat-sen University, Guangzhou 510080, China
²Department of Urology, Huadu District People’s Hospital of Guangzhou, Guangzhou 510000, China
³Department of Urology, Tianjin Medical University Cancer Institute & Hospital, Tianjin 300060, China
⁴Department of Urology, the First Affiliated Hospital of Guangzhou Medical University, Guangzhou 510030, China

Received:2025-08-29 Published:2026-08-01
Corresponding author: Junhang Luo

全文 ( ) 下载PDF ( ) 导出引用

引用本文：

李佳颖, 刘溪, 陈宇航, 桂程鹏, 习明, 陈炜, 罗俊航. 卷积神经网络在预测肾透明细胞癌全视野数字切片肿瘤坏死中的应用研究[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2026, 20(04): 389-398.

Jiaying Li, Xi Liu, Yuhang Chen, Chengpeng Gui, Ming Xi, Wei Chen, Junhang Luo. Application of convolutional neural networks in predicting tumor necrosis from whole-slide images of clear cell renal cell carcinoma[J/OL]. Chinese Journal of Endourology(Electronic Edition), 2026, 20(04): 389-398.

目的

建立卷积神经网络模型预测肾透明细胞癌（ccRCC）全视野数字切片（WSI）肿瘤坏死，评估模型预测效果，并量化坏死比例。

方法

下载TCGA数据库25张存在肿瘤坏死的ccRCC WSI，扫描中山大学附属第一医院100张ccRCC病理玻片，在病理医师指导下对肿瘤及坏死区域进行标注。将TCGA来源的WSI切割为21 411张图像块（2 556张坏死，18 855张非坏死），对卷积神经网络DenseNet201进行训练，建立检测肿瘤及坏死图像块的深度学习模型，并使用本中心来源的3 855张图像块进行外部测试。根据图像块分类结果和位置信息，制作肿瘤坏死分布预测图，运用设定分位数阈值的方法分析模型在整体切片水平检测肿瘤坏死的能力。

结果

在外部测试中，模型在图像块水平分类肿瘤/非肿瘤准确度为0.909，灵敏度为0.890，特异度为0.943，受试者工作特征曲线下面积（AUC）为0.973；分类坏死/非坏死准确度为0.717，灵敏度为0.958，特异度为0.657，AUC为0.920。在切片水平检测肿瘤坏死的AUC为0.903，灵敏度为0.760，特异度为0.933，准确度为0.890。根据图像块分类结果生成肿瘤坏死预测图，并计算肿瘤坏死百分数。

结论

基于卷积神经网络算法的分类模型可以作为一种辅助诊断工具，帮助病理医师检测ccRCC肿瘤坏死并对坏死范围进行量化评估。

关键词: 肾透明细胞癌, 肿瘤坏死, 卷积神经网络, 全视野数字切片, 人工智能

Objective

To build a convolutional neural network model to predict tumor necrosis on whole-slide images (WSI) in clear cell renal cell carcinoma, and to assess the performance of the model.

Methods

25 WSIs of clear cell renal cell carcinoma with tumor necrosis are downloaded from TCGA database, and 100 pathological slides of clear cell renal cell carcinoma from the First Affiliated Hospital of Sun Yat-sen University are scanned. Tumor and necrosis region are annotated in WSIs under the guidance of pathologists. The WSIs from TCGA are then cropped into 21 411 patches (2 556 necrosis and 18 855 non-necrosis), training the convolutional neural network DenseNet201 to establish a deep learning model for detecting tumor and necrotic patches. 3 855 patches from our center are used for external testing. Based on the classification results and location information of patches, tumor necrosis-prediction maps are generated. The quantile threshold method is used to analyze the performance of the model to detect tumor necrosis at the whole slide level.

Results

In the external testing group, the accuracy, sensitivity, specificity and AUC of the model at patch-wise are 0.909, 0.890, 0.943, and 0.973 for tumor and non-tumor classification. The accuracy, sensitivity, specificity and AUC are 0.717, 0.958, 0.657 and 0.920 respectively for necrotic and non-necrotic classification. The AUC, sensitivity, specificity and accuracy of tumor necrosis at slide-wise are 0.903, 0.760, 0.933, and 0.890. Based on the classification results of patches, prediction maps of tumor necrosis are generated and the percentage of tumor necrosis is calculated.

Conclusion

The classification model based on convolutional neural network algorithm can be used as an assistant diagnostic tool to help pathologists detect tumor necrosis clear cell renal cell carcinoma and quantitatively evaluate the extent of necrosis.

Key words: Clear cell renal cell carcinoma, Tumor necrosis, Convolutional Neural Networks, Whole-slide images, Artificial intelligence

图1 卷积神经网络在预测肾透明细胞癌全视野数字切片（WSI）肿瘤坏死中的应用研究流程图

图2 本研究数据集构建方法注：a为将下载的WSI图像放大至40×倍数水平；b为切割为无重叠区域的2 048×2 048像素的图像块；c为剔除空白或污染区域>80%的图像块；d为得到可用于训练模型的图像块

图3 肾透明细胞癌WSI肿瘤坏死表现注：a、b为10倍、40倍镜下颗粒样坏死，即边界清晰的坏死灶及丰富的胞质胞核碎片；c、d为10倍、40倍镜下凝固性坏死，即死亡和降解的肿瘤细胞以簇状和片状合并成嗜伊红的无定形凝固物，其中无核碎片；e、f为10倍、40倍镜下脏坏死，即存在大量炎症细胞及细胞碎片

图4 卷积神经网络DenseNet201模型结构示意图注：DenseNet201由卷积层、池化层、全连接层以及特有的稠密区块构成

图5 TCGA数据库来源肾透明细胞癌WSI图像块分布注：TCGA数据库来源的25张肾透明细胞癌WSI中，a为肿瘤/非肿瘤组得到21 270张图像块，其中16 753张为肾透明细胞癌组织，4 517张为非肿瘤组织；b为坏死/非坏死组得到21 411张图像块，其中2 556张为肿瘤坏死组织，18 855张为非坏死组织

图6 TCGA数据库来源内部测试集图像块预测肿瘤及坏死概率值注：按照真实分类进行分组，a为肿瘤组平均概率值为0.882，非肿瘤组平均概率值为0.342，两组概率值差异有统计学意义（P<0.001）；b为坏死组平均概率值为0.719，非坏死组平均概率值为0.342，两组概率值差异有统计学意义（P<0.001）

图7 DenseNet201模型图像块层面分类肿瘤/非肿瘤和坏死/非坏死的ROC曲线注：a为内部测试集在分类肿瘤/非肿瘤任务中，AUC为0.903（0.95CI：0.892~0.914）；b为内部测试集在分类坏死/非坏死任务中，AUC为0.932（0.95CI：0.925~0.939）；c为外部测试集在分类肿瘤/非肿瘤任务中，AUC为0.973（0.95CI：0.968~0.977）；d为外部测试集在分类坏死/非坏死任务中，AUC为0.920（0.95CI：0.912~0.929）

表1 WSI切割后图像块层面分类肿瘤/非肿瘤和坏死/非坏死

组别	肿瘤/非肿瘤				坏死/非坏死
组别	AUC	准确度	灵敏度	特异度	AUC	准确度	灵敏度	特异度
内部测试	0.903	0.781	0.889	0.675	0.932	0.851	0.890	0.850
外部测试	0.973	0.909	0.890	0.943	0.920	0.717	0.958	0.657

表1 WSI切割后图像块层面分类肿瘤/非肿瘤和坏死/非坏死

组别	肿瘤/非肿瘤				坏死/非坏死
组别	AUC	准确度	灵敏度	特异度	AUC	准确度	灵敏度	特异度
内部测试	0.903	0.781	0.889	0.675	0.932	0.851	0.890	0.850
外部测试	0.973	0.909	0.890	0.943	0.920	0.717	0.958	0.657

图8 DenseNet201模型对外部测试集图像块三种坏死类型预测能力分析注：a为颗粒样坏死预测结果混淆矩阵，b为凝固性坏死预测结果混淆矩阵，c为脏坏死预测结果混淆矩阵；模型预测三种坏死的灵敏度分别为0.967、0.975、0.874，准确度分别为0.692、0.684、0.664

图9 DenseNet201模型肿瘤坏死预测图注：a为WSI原图，b为病理医师分类图像块的结果，c为DenseNet201模型分类结果，红色代表坏死区域，绿色代表非坏死的肿瘤区域，蓝色代表非肿瘤区域

图10 病理医师分类与DenseNet201模型预测肿瘤坏死的相关性

图11 切片层面（每张WSI所有图像块坏死预测概率值）相关参数分类坏死/非坏死ROC曲线注：采用坏死预测概率值中位数、90%分位数、99%分位数、平均数对于切片进行分类，其中99%分位数分类的AUC最高，为0.903（0.95CI：0.828~0.978）

图12 本中心100张WSI 99%分位预测概率值注：坏死组99%分位平均概率值为0.992，非坏死组99%分位数平均概率值为0.866，两组99%分位概率值差异有统计学意义（P<0.001）

[1]	Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024 [J]. CA Cancer J Clin, 2024, 74(1): 12-49. DOI: 10.3322/caac.21820.
[2]	Bex A., Abu-Ghanem Y., Bedke J., et al. EAU guidelines on renal cell carcinoma[EB/OL].2026. URL
[3]	Zhang L, Zha Z, Qu W, et al. Tumor necrosis as a prognostic variable for the clinical outcome in patients with renal cell carcinoma: a systematic review and meta-analysis[J]. BMC Cancer, 2018, 18(1): 870. DOI: 10.1186/s12885-018-4773-z.
[4]	钱洲楠, 邹元章, 陈兵海. 通过生物信息学分析鉴定肾透明细胞癌中的候选核心基因及其与免疫细胞浸润的关系[J]. 江苏大学学报(医学版), 2025, 35(6): 512-520, 527. DOI: 10.13312/j.issn.1671-7783.y240224.
[5]	Leibovich BC, Blute ML, Cheville JC, et al. Prediction of progression after radical nephrectomy for patients with clear cell renal cell carcinoma: a stratification tool for prospective clinical trials[J]. Cancer, 2003, 97(7): 1663-1671. DOI: 10.1002/cncr.11234.
[6]	Delahunt B, McKenney JK, Lohse CM, et al. A novel grading system for clear cell renal cell carcinoma incorporating tumor necrosis[J]. Am J Surg Pathol, 2013, 37(3): 311-322. DOI: 10.1097/PAS.0b013e318270f71c.
[7]	Klatte T, Said JW, de Martino M, et al. Presence of tumor necrosis is not a significant predictor of survival in clear cell renal cell carcinoma: higher prognostic accuracy of extent based rather than presence/absence classification[J]. J Urol, 2009, 181(4): 1558-1564. DOI: 10.1016/j.juro.2008.11.098.
[8]	Katz MD, Serrano MF, Grubb RL 3rd, et al. Percent microscopic tumor necrosis and survival after curative surgery for renal cell carcinoma[J]. J Urol, 2010, 183(3): 909-914. DOI: 10.1016/j.juro.2009.11.010.
[9]	Minardi D, Lucarini G, Filosa A, et al. Prognostic role of tumor necrosis, microvessel density, vascular endothelial growth factor and hypoxia inducible factor-1alpha in patients with clear cell renal carcinoma after radical nephrectomy in a long term follow-up[J]. Int J Immunopathol Pharmacol, 2008, 21(2): 447-455. DOI: 10.1177/039463200802100225.
[10]	Delahunt B, Cheville JC, Martignoni G, et al. The international society of urological pathology (ISUP) grading system for renal cell carcinoma and other prognostic parameters[J]. Am J Surg Pathol, 2013, 37(10): 1490-1504. DOI: 10.1097/PAS.0b013e318299f0fb.
[11]	Awan R, Sirinukunwattana K, Epstein D, et al. Glandular morphometrics for objective grading of colorectal adenocarcinoma histology images[J]. Sci Rep, 2017, 7(1): 16852. DOI: 10.1038/s41598-017-16516-w.
[12]	Chen M, Zhang B, Topatana W, et al. Classification and mutation prediction based on histopathology H&E images in liver cancer using deep learning[J]. NPJ Precis Oncol, 2020, 4: 14. DOI: 10.1038/s41698-020-0120-3.
[13]	Litjens G, Sánchez CI, Timofeeva N, et al. Deep learning as a tool for increased accuracy and efficiency of histopathological diagnosis[J]. Sci Rep, 2016, 6: 26286. DOI: 10.1038/srep26286.
[14]	Wang S, Zhu Y, Yu L, et al. RMDL: Recalibrated multi-instance deep learning for whole slide gastric image classification[J]. Med Image Anal, 2019, 58: 101549. DOI: 10.1016/j.media.2019.101549.
[15]	Samaratunga H, Delahunt B, Srigley JR, et al. Granular necrosis: a distinctive form of cell death in malignant tumours[J]. Pathology, 2020, 52(5): 507-514. DOI: 10.1016/j.pathol.2020.06.002.
[16]	Sengupta S, Lohse CM, Leibovich BC, et al. Histologic coagulative tumor necrosis as a prognostic indicator of renal cell carcinoma aggressiveness[J]. Cancer, 2005, 104(3): 511-520. DOI: 10.1002/cncr.21206.
[17]	Kuroe T, Watanabe R, Kojima M, et al. Evaluation of the morphological features and unfavorable prognostic impact of dirty necrosis in renal cell carcinoma[J]. J Cancer Res Clin Oncol, 2021, 147(4): 1089-1100. DOI: 10.1007/s00432-020-03505-2.
[18]	Huang G, Liu Z, Pleiss G, et al. Convolutional networks with dense connectivity[J]. IEEE Trans Pattern Anal Mach Intell, 2022, 44(12): 8704-8716. DOI: 10.1109/TPAMI.2019.2918284.
[19]	Dagher J, Delahunt B, Rioux-Leclercq N, et al. Assessment of tumour-associated necrosis provides prognostic information additional to World Health Organization/International Society of Urological Pathology grading for clear cell renal cell carcinoma[J]. Histopathology, 2019, 74(2): 284-290. DOI: 10.1111/his.13737.
[20]	Cruz-Roa AA, Arevalo Ovalle JE, Madabhushi A, et al. A deep learning architecture for image representation, visual interpretability and automated basal-cell carcinoma cancer detection[J]. Med Image Comput Comput Assist Interv, 2013, 16(Pt 2): 403-410. DOI: 10.1007/978-3-642-40763-5_50.
[21]	Cruz-Roa AA, Basavanhally A, González F, et al. Automatic detection of invasive ductal carcinoma in whole slide images with convolutional neural networks[J]. Med Imag 2014 Digit Pathol, 2014, 9041: 904103. DOI: 10.1117/12.2043872.
[22]	Araújo T, Aresta G, Castro E, et al. Classification of breast cancer histology images using convolutional neural networks[J]. PLoS One, 2017, 12(6): e0177544. DOI: 10.1371/journal.pone.0177544.
[23]	Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning[J]. Nat Med, 2018, 24(10): 1559-1567. DOI: 10.1038/s41591-018-0177-5.
[24]	Wei JW, Suriawinata AA, Vaickus LJ, et al. Evaluation of a deep neural network for automated classification of colorectal polyps on histopathologic slides[J]. JAMA Netw Open, 2020, 3(4): e203398. DOI: 10.1001/jamanetworkopen.2020.3398.
[25]	Tabibu S, Vinod PK, Jawahar CV. Pan-renal cell carcinoma classification and survival prediction from histopathology images using deep learning[J]. Sci Rep, 2019, 9(1): 10509. DOI: 10.1038/s41598-019-46718-3.
[26]	Fenstermaker M, Tomlins SA, Singh K, et al. Development and validation of a deep-learning model to assist with renal cell carcinoma histopathologic interpretation[J]. Urology, 2020, 144: 152-157. DOI: 10.1016/j.urology.2020.05.094.
[27]	Arunachalam HB, Mishra R, Daescu O, et al. Viable and necrotic tumor assessment from whole slide images of osteosarcoma using machine-learning and deep-learning models[J]. PLoS One, 2019, 14(4): e0210706. DOI: 10.1371/journal.pone.0210706.
[28]	Sharma H, Zerbe N, Klempert I, et al. Deep convolutional neural networks for automatic classification of gastric carcinoma using whole slide images in digital histopathology[J]. Comput Med Imaging Graph, 2017, 61: 2-13. DOI: 10.1016/j.compmedimag.2017.06.001.
[29]	Avulova S, Cheville JC, Lohse CM, et al. Grading chromophobe renal cell carcinoma: evidence for a four-tiered classification incorporating coagulative tumor necrosis[J]. Eur Urol, 2021, 79(2): 225-231. DOI: 10.1016/j.eururo.2020.10.007.
[30]	Parker WP, Cheville JC, Frank I, et al. Application of the stage, size, grade, and necrosis (SSIGN) score for clear cell renal cell carcinoma in contemporary patients[J]. Eur Urol, 2017, 71(4): 665-673. DOI: 10.1016/j.eururo.2016.05.034.
[31]	Pichler M, Hutterer GC, Chromecki TF, et al. Histologic tumor necrosis is an independent prognostic indicator for clear cell and papillary renal cell carcinoma[J]. Am J Clin Pathol, 2012, 137(2): 283-289. DOI: 10.1309/AJCPLBK9L9KDYQZP.
[32]	Pichler M, Hutterer GC, Chromecki TF, et al. Presence and extent of histological tumour necrosis is an adverse prognostic factor in papillary type 1 but not in papillary type 2 renal cell carcinoma[J]. Histopathology, 2013, 62(2): 219-228. DOI: 10.1111/j.1365-2559.2012.04328.x.
[33]	Zini L, Leroy X, Lemaitre L, et al. Tumour necrosis in chromophobe renal cell carcinoma: clinical data to distinguish aggressive variants[J]. Eur J Surg Oncol, 2008, 34(6): 687-691. DOI: 10.1016/j.ejso.2007.08.010.

[1]	高嘉敏, 韩俊毅. 老年胃肠癌患者多模式预康复：异质性挑战与分类管理策略[J/OL]. 中华普通外科学文献(电子版), 2026, 20(03): 192-199.
[2]	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.
[3]	许颜浩, 邓正根, 张晓坡, 甘卫东, 郭宏骞. 基于CT角度界面和溢啤酒征对乏脂性肾血管平滑肌脂肪瘤与肾透明细胞癌的鉴别[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2026, 20(03): 273-278.
[4]	石亚超, 魏六木, 东小鸽, 樊海宁, 侯立朝, 杜凯豪, 汪占金, 薛伟伟, 刘海刚, 王展. 肝癌多模式诊断研究进展与展望[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(03): 326-336.
[5]	邓玉飞, 王志鑫, 娄珂, 张林轩, 马桂春, 港措. 影像组学在肝癌精准诊断、疗效评估及治疗方案决策优化中应用[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(02): 172-180.
[6]	杨春元, 邓旭, 王晶晶, 阳丹才让, 潘伟. 精准肝切除术技术进展与临床应用[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(02): 261-265.
[7]	崔妍, 祖盼云, 宋宇, 谢克亮. 基于人工智能的线上案例平台在研究生重症医学教学中的应用[J/OL]. 中华重症医学电子杂志, 2026, 12(02): 172-179.
[8]	钱涛. 融合与重塑：神经调控康复一体化范式的构建与展望[J/OL]. 中华脑科疾病与康复杂志(电子版), 2026, 16(02): 65-70.
[9]	秦士钊, 刘晓. 消化性溃疡患者血清CD4⁺ T细胞、miR-26a和TNF-α水平特征及与幽门螺杆菌感染的关系[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(03): 243-247.
[10]	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.
[11]	余绍斌, 康明强. ⅠA3期非小细胞肺癌预后差异及治疗策略研究进展[J/OL]. 中华胸部外科电子杂志, 2026, 13(02): 132-139.
[12]	刘诗馨, 宋晓微, 武剑. 临床视角下人工智能赋能脑卒中康复的范式重塑[J/OL]. 中华脑血管病杂志(电子版), 2026, 20(03): 231-239.
[13]	梁怡凡, 牟婧宇, 吴雅婷, 邳靖陶, 陈乐, 武剑. 不同人工智能模型预测脑卒中不良预后诊断效能的荟萃分析[J/OL]. 中华脑血管病杂志(电子版), 2026, 20(03): 308-319.
[14]	张思远, 宋晓微, 王丽君, 范玉华, 武剑. 人工智能赋能脑小血管病的血流动力学评估：研究进展与临床转化[J/OL]. 中华脑血管病杂志(电子版), 2026, 20(03): 327-333.
[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

选择文件类型/文献管理软件名称

选择包含的内容

Application of convolutional neural networks in predicting tumor necrosis from whole-slide images of clear cell renal cell carcinoma

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

Application of convolutional neural networks in predicting tumor necrosis from whole-slide images of clear cell renal cell carcinoma