改进残差网络的肾细胞癌ISUP分级研究

doi:10.3778/j.issn.1002-8331.2208-0279

摘要/Abstract

摘要： 术前预测透明细胞肾细胞癌（clear cell renal cell carcinoma，ccRCC）的分级可有效评估患者的预后并指导临床治疗，但实现精准预测是目前本领域内的一项重要问题。该研究首先确定最优建模的CT类型与网络层数，提出了一种基于改进残差网络的ccRCC的CT影像分级模型，具体包括：利用大卷积操作对图像进行原始特征提取，利用混合注意力模块通过计算特征图中当前空间和临近空间以及当前空间和远距离空间之间的信息交互获取更多有用的特征，使得原始图像特征图在通道维度与空间维度上进行自适应特征细化，利用四个深度卷积网络层提取图像深度特征，并利用改进通道注意力模块产生通道注意力特征图信息，提取更多通道上的交互信息。实验结果表明，增强CT实质期图像和34层残差网络最有利于分级预测模型的开发，所提出的模型的总体加权准确率、AUC、精度、召回率和F1分数分别为90.8%、0.897、90.5%、90.8%、90.9%，各项指标优于其他常见网络结构，因此，该模型在预测ccRCC的国际泌尿病理学学会（International Society of Urological Pathology，ISUP）分级方面有良好的效能，对患者的临床辅助诊断和预后治疗具有重要的理论指导意义。

关键词: 透明细胞肾细胞癌（ccRCC）, ISUP分级, CT, 增强CT实质期, 通道注意力, 空间注意力

Abstract: Clear cell renal cell carcinoma (ccRCC) grading prior to surgery is of great importance for effectively evaluating the prognosis of patients or guiding clinical treatment. But the realization of accurate prediction is an important problem in the field. The optimal computed tomography (CT) type and network layer number for modeling are determined, and a CT image grading model based on ccRCC with improved residual network is proposed in this paper. Firstly, the model uses large convolution operation to extract original features from the images. Then, the improved module calculates the information interaction of the current space and local space, and calculates additionally the interaction of the current space and remote space. The original image features are adaptively refined in channel dimension and spatial dimension. Then, improved channel attention module makes the depth feature generate channel attention feature map information. Interaction information on more channels is extracted. The experimental results show that the enhanced parenchyma CT images and 34-layer residual network are most beneficial to the development of grading model. The indexes of the proposed model are better than other common network structures. The overall weighted accuracy reaches 90.8%, the AUC and precision reach 0.897 and 90.5%, the recall and F1-score reach 90.8% and 90.9%. Therefore, the deep learning model has good performance in predicting International Society of Urological Pathology (ISUP) grade of ccRCC, and has important theoretical guiding significance for clinical auxiliary diagnosis and prognostic treatment of patients.

Key words: clear cell renal cell carcinoma (ccRCC), International Society of Urological Pathology(ISUP) grade, CT, enhanced CT parenchymal phase, channel attention, space attention

孙振铎, 张明洋, 王向辉, 赵磊, 刘爽, 崔振宇, 杨昆, 刘琨. 改进残差网络的肾细胞癌ISUP分级研究[J]. 计算机工程与应用, 2024, 60(2): 221-230.

SUN Zhenduo, ZHANG Mingyang, WANG Xianghui, ZHAO Lei, LIU Shuang, CUI Zhenyu, YANG Kun, LIU Kun. Research on ISUP Grade of Renal Cell Carcinoma by Improved ResNet Network[J]. Computer Engineering and Applications, 2024, 60(2): 221-230.

参考文献

[1] SUNG H, FERLAY J, SIEGEL R L, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA, 2021, 71(3): 209-249.
[2] SIEGEL R L, MILLER K D, FUCHS H E, et al. Cancer statistics, 2022[J]. CA, 2022, 72(1): 7-33.
[3] SRIGLEY J R, DELAHUNT B, EBIE J N, et al. The International Society of Urological Pathology (ISUP) Vancouver classification[J]. American Journal of Surgical Pathology, 2013, 37(10): 1469-1489.
[4] JEWETT M A S, MATTAR K, BASIUK J, et al. Active surveillance of small renal masses: progression patterns of early stage kidney cancer[J]. European Urology, 2011, 60(1): 39-44.
[5] BEKTAS C T, KOCAK B, YARDIMCI A H, et al. Clear cell renal cell carcinoma: machine learning-based quantitative computed tomography texture analysis for prediction of Fuhrman nuclear grade[J]. European Radiology, 2019, 29(3): 1153-1163.
[6] KOCAK B, DURMAZ E S, ATES E, et al. Unenhanced CT texture analysis of clear cell renal cell carcinomas: a machine learning-based study for predicting histopathologic nuclear grade[J]. American Journal of Roentgenology, 2019, 212(6) : 132-139.
[7] MA Y Q, GUAN Z, LIANG H, et al. Predicting the WHO/ISUP grade of clear cell renal cell carcinoma through CT-based tumoral and peritumoral radiomics[J]. Frontiers in Oncology, 2022, 12: 831112.
[8] ZHOU L, ZHANG Z H, CHEN Y H, et al. A deep learning-based radiomics model for differentiating benign and malignant renal tumors[J]. Translational Oncology, 2019, 12(2): 292-300.
[9] BYUN S S, HEO T S, CHOI J M, et al. Deep learning based prediction of prognosis in nonmetastatic clear cell renal cell carcinoma[J]. Scientific Reports, 2021, 11(1): 1242.
[10] LIN F, MA C Y, XU J P, et al. A CT-based deep learning model for predicting the nuclear grade of clear cell renal cell carcinoma[J]. European Journal of Radiology, 2020, 129: 109079.
[11] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[12] 黄英来, 艾昕. 改进残差网络在玉米叶片病害图像的分类研究[J]. 计算机工程与应用, 2021, 57(23): 178-184.
HUANG Y L, AI X. Research on classification of corn leaf disease image by improved residual network[J]. Computer Engineering and Applications, 2021, 57(23): 178-184.
[13] 林雍博, 凌捷. 基于残差网络和GRU的XSS攻击检测方法[J]. 计算机工程与应用, 2022, 58(10): 101-107.
LIN Y B, LIN J. XSS attack detection method based on residual network and GRU[J]. Computer Engineering and Applications, 2022, 58(10): 101-107.
[14] JIANG J, ELGUINDI S, BERRY S, et al. Nested-block self-attention multiple resolution residual network for multi-organ segmentation from CT[J]. Medical Physics, 2022, 49(8): 5244-5257.
[15] SZEGDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[J]. arXiv:1409.4842,2014.
[16] HOWARD A, SANDLER M, CHU G, et al. Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision, 2019: 1314-1324.
[17] HAKIMI A A, PHAM C G, HSIEH J J. A clear picture of renal cell carcinoma[J]. Nature Genetics, 2013, 45(8): 849-850.
[18] STRANSKY L A, VLGEANT S M, HUANG B, et al. Sensitivity of VHL mutant kidney cancers to HIF2 inhibitors does not require an intact p53 pathway[J]. Proceedings of the National Academy of Sciences of the United States of America, 2022, 119(14): 2120403119.
[19] GOSSAGE L, EISEN T, MAHER E R. VHL, the story of a tumour suppressor gene[J]. Nature Reviews Cancer, 2015, 15(1): 55-64.
[20] SEMENZA G L. HIF-1 mediates metabolic responses to intratumoral hypoxia and oncogenic mutations[J]. Journal of Clinical Investigation, 2013, 123(9): 3664-3671.
[21] HSIEH J J, LE V, CAO D F, et al. Genomic classifications of renal cell carcinoma: a critical step towards the future application of personalized kidney cancer care with pan-omics precisio[J]. The Journal of pathology, 2018, 244(5): 525-537.