醫用超音波影像是慢性腎臟病(Chronic Kidney Disease, CKD)診斷的重要工具，雖然超音波影像判讀經常高度依賴臨床醫師的主觀經驗，但人工智能(Artificial Intelligence, AI)利用卷積神經網路(Convolutional Neural Networks, CNNs)可輔助醫師提升對超音波影像判讀的客觀性。本論文回溯性蒐集了三間醫院過去十年內的腎臟超音波影像與切片病理報告以實現AI利用超音波影像對腎臟間質纖維化與腎小管萎縮(Interstitial fibrosis and tubular atrophy, IFTA)程度的預測，我們會先將含有完整腎臟且不具有假影的超音波影像輸入至Mask R-CNN模型進行ROI提取以降低背景對於預測結果的干擾，並且以重疊度(Intersection over Union, IoU)以及Dice係數作為分析指標，再經過我們提出的雙路徑卷積神經網路(Dual-Path Convolutional Neural Network, DPCNN)同時提取並整合影像中的高階特徵與低階特徵來進行IFTA的預測，該DPCNN模型在5倍交叉驗證下針對IFTA程度之二分類預測表現以及95%信賴區間得到了正確率0.856 (0.818-0.876)、召回率0.761 (0.715-0.817)、特異性0.927 (0.862-0.952)、精確率0.887 (0.804-0.920)、F1分數0.819 (0.776-0.846)、受試者工作特徵曲線下面積0.922 (0.893-0.944)，其結果均明顯優於其他現有卷積神經網路模型。

Medical ultrasound (US) imaging is an important tool for the clinical diagnosis of chronic kidney disease (CKD). Although the interpretation of US images highly depends on the subjective experience of clinicians, artificial intelligence (AI) using convolutional neural networks (CNN) can assist clinicians to improve their objectivity. In this study, we retrospectively collected the renal US images and biopsy pathology reports from three hospitals in the past ten years to realize the prediction of the stage of renal interstitial fibrosis and tubular atrophy (IFTA) by AI-assisted interpretation of US images. We will first input the renal US images with intact kidneys and no artifacts into the Mask R-CNN model for ROI extraction with Intersection over Union (IoU) and Dice coefficient as quantitative metrics. Then, the proposed dual-path convolutional neural network (DPCNN) is used to simultaneously extract and integrate high-level and low-level features in the US image for IFTA prediction. With five-fold cross-validation, the proposed DPCNN for binary IFTA classification achieves the accuracy of 0.856 (0.818-0.876), the recall of 0.761 (0.715-0.817), the specificity of 0.927 (0.862-0.952), the precision of 0.887 (0.804-0.920), the F1 score of 0.819 (0.776-0.846) and the area under the receiver operating characteristic curve (AUC) of 0.922 (0.893-0.944). The results are all significantly better than other existing convolutional neural network models.

[1] 110年死因統計結果分析，衛生福利部統計處，民國111年。
[2] 2020台灣腎病年報，財團法人國家衛生研究院&台灣腎臟醫學會，民國110年。
[3] Chia-Tien Hsu, Chun-Te Huang, Shang-Feng Tsai, Ming-Ju Wu, Cheng-Hsu Chen, “Acute Kidney Injury-Chronic Kidney Disease Continuum,” Kidney and Dialysis, Vol.32, pp17-pp21, 2020.
[4] “USRD 2020 Annual Report,” USRDS Coordinating Center, 2021.
[5] 鍾國亮，影像處理與電腦視覺7e，東華書局，2020。
[6] K. He, G. Gkioxari, P. Dollar, and R. Girshick, “Mask R-CNN,” in IEEE International Conference on Computer Vision (ICCV), 2017.
[7] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” in Neural Information Processing Systems (NIPS), 2015.
[8] J. Long, E. Shelhamer, and T. Darrell, “Fully Convolutional Networks for Semantic Segmentation,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
[9] T.-Y. Lin, P. Dollar, R. Girshick, K. He, B. Hariharan, and S. Belongie. “Feature Pyramid Networks for Object Detection,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.
[10] R. Girshick, “Fast R-CNN,” in IEEE International Conference on Computer Vision (ICCV), 2015.
[11] Z. Liu et al. “Diagnosis of significant liver fibrosis in patients with chronic hepatitis B using a deep learning-based data integration network,” Hepatol Int 16, 526–536, 2022.
[12] Q. Liu, Z. Liu, W. Xu, H. Wen, M. Dai and X. Chen, “Diagnosis of Significant Liver Fibrosis by Using a DCNN Model With Fusion of Features From US B-Mode Image and Nakagami Parametric Map: An Animal Study,” in IEEE Access, vol. 9, pp. 89300-89310, 2021.
[13] K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” in International Conference on Learning Representations (ICLR), 2015.
[14] Y. Hu, B. Xia, M. Mao, Z. Jin, J. Du, L.Guo, and A. F. Frangi, “AIDAN: An Attention-Guided Dual-Path Network for Pediatric Echocardiography Segmentation,” IEEE Access., vol. 8, pp. 29176-29187, 2020.
[15] Derchi LE, Martinoli C, Saffioti S, Pontremoli R, De Micheli A, Bordone C, “Ultrasonographic imaging and Doppler analysis of renal changes in non-insulin-dependent diabetes mellitus. Acad Radiol,” 1994.
[16] Buturović-Ponikvar J, Visnar-Perovic A, “Ultrasonography in chronic renal failure,” Eur J Radiol, 2003.
[17] S. Sethi, V.D. D’Agati, C.C. Nast, A.B. Fogo, A.S. De Vriese, G.S. Markowitz, et al. “A proposal for standardized grading of chronic changes in native kidney biopsy specimens,” Kidney International, Volume 91, Issue 4, pp. 787-789, 2017.
[18] K. Zuiderveld, “Contrast Limited Adaptive Histogram Equalization,” Chapter VIII.5, Graphics Gems IV. P.S. Heckbert (Eds.), Cambridge, MA, Academic Press, pp. 474–485, 1994.
[19] M. D. Cosmo, M. C. Fiorentino, F. P. Villani, G. Sartini, G. Smerilli, E. Filippucci, E. Frontoni, S. Moccia, “Learning-Based Median Nerve Segmentation From Ultrasound Images For Carpal Tunnel Syndrome Evaluation,” Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2021.
[20] G. Kompella, M. Antico, F. Sasazawa, S. Jeevakala, K. Ram, D. Fontanarosa, A. K Pandey, M. Sivaprakasam, “Segmentation of Femoral Cartilage from Knee Ultrasound Images Using Mask R-CNN,” Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2019.
[21] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 770–778, 2016.
[22] A. Marcu and M. Leordeanu, “Dual local-global contextual pathways for recognition in aerial imagery,” May 2016.
[23] J. Hu, L. Shen, G. Sun, “Squeeze-and-excitation networks,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.
[24] S. Ioffe, C. Szegedy, “Batch Normalization Accelerating Deep Network Training by Reducing Internal Covariate Shift,” International Conference on Machine Learning. pp. 448-456, 2015.
[25] C. Yu, J. Wang, C. Peng, C. Gao, G. Yu and N. Sang, “BiSeNet: Bilateral segmentation network for real-time semantic segmentation,” Computer Vision—ECCV, vol. 11217, pp. 334-349, 2018.
[26] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-cam: Visual explanations from deep networks via gradient-based localization,” in International Conference on Computer Vision (ICCV), 2017.
[27] T. Y. Lin, M. Maire, S. Belongie, L. Bourdev and R. Girshick. (2015). “Microsoft COCO: Common Objects in Context,” [Online]. Available: https://arxiv.org/abs/1405.0312.
[28] T. Fawcett, “An introduction to ROC analysis,” Pattern Recognit. Lett.,vol. 27, no. 8, pp. 861-874, Jun. 2006.
[29] C. J. Clopper and E. S. Pearson, “The use of confidence or fiducial limits illustrated in the case of the binomial,” Biometrika, vol. 26, no. 4, pp. 404-413, 1934.
[30] E. R. DeLong, D. M. DeLong, and D. L. Clarke-Pearson, “Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach,” Biometrics, vol. 44, pp. 837 845, Sep. 1988.
[31] C. Szegedy, V. Vanhoucke, S. Ioffe and J. Shlens. (2015). “Rethinking the Inception Architecture for Computer Vision,” [Online]. Available: https://arxiv.org/abs/1512.00567.
[32] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov and L.C. Chen. (2019). “MobileNetV2: Inverted Residuals and Linear Bottlenecks” [Online]. Available: https://arxiv.org/abs/1801.04381.
[33] G. Huang, Z. Liu, L. van der Maaten, K. Q. Weinberger. (2018). “Densely Connected Convolutional Networks” [Online]. Available: https://arxiv.org/abs/1608.06993.
[34] J. Deng, W. Dong, R. Socher, L. J. Li, K. Li, F. F. Li, “ImageNet: A Large-Scale Hierarchical Image Database,” in Conference on Computer Vision and Pattern Recognition (CVPR), 2009.