Basic Search / Detailed Display

Author: 張海川
Hai-Chuan Zhang
Thesis Title: 基於雙輸出網路的胎兒頭部超音波影像分割測量方法
Ultrasound Fetal Head Segmentation and Measurement Method Using Dual Output Networks
Advisor: 洪西進
Shi-Jinn Horng
Committee: 謝仁偉
Jen-Wei Hsieh
李正吉
Cheng-Chi Lee
林祝興
Chu-Hsing Lin
吳怡樂
Yi-Leh Wu
楊竹星
Chu-Sing Yang
Degree: 碩士
Master
Department: 電資學院 - 資訊工程系
Department of Computer Science and Information Engineering
Thesis Publication Year: 2022
Graduation Academic Year: 110
Language: 中文
Pages: 56
Keywords (in Chinese): 深度學習神經網絡圖像分割
Keywords (in other languages): Deep learning, Neural networks, Image segmentation
Reference times: Clicks: 210Downloads: 7
Share:
School Collection Retrieve National Library Collection Retrieve Error Report

隨着高效能GPU在醫療設備上的廣泛運用和醫用超音波影像設備的更新換代,越來越多的醫用超音波診斷設備開始利用深度學習技術完成圖像分類、圖像分割和目標檢測等任務。對胎兒頭部超音波影像進行自動測量,在過去常利用形態學和傳統統計學的算法,常存在結果不準確和執行時間長的問題。現階段已有較多產品將深度學習的圖像分割技術用於胎兒頭部整體區域測量。然而在產科臨床測量中,不僅需識別出超音波影像中胎兒頭圍區域以計算測定胎兒發育所需的枕額徑(OFD),也需估算出胎兒顱骨的厚度以求得另一重要指標雙頂徑(BPD)。利用現有的深度學習測量方法,只能在分割出胎兒頭部區域後再利用傳統的形態學或統計學運算取得顱骨區域並估算其厚度,相較一般的端到端方法增加大量運算時間。同時在公開的胎兒頭部影像分割資料集中,幾乎未見有提供對於胎兒顱骨區域的分割標註,這種情況也增加了利用深度學習分割模型完成胎兒頭圍和顱骨區域端到端自動測量的困難程度。

本研究提出了一個基於Deeplab V3 Plus的雙輸出網路用於實現胎兒頭部超音波影像的頭圍、顱骨端到端分割方法,並提出了利用傳統形態學方法生成並製作胎兒顱骨分割標註資料集的方法解決上述問題。本研究對胎兒頭圍和顱骨的型態結構關聯性提出相互促進的LOLI Loss,幫助網路實現更高的分割準確度。本研究亦對當前流行的多種骨幹網路與Attention機制進行實驗,以探討這些骨幹網路和Attention機制對於卷積神經網絡的胎兒頭圍-顱骨分割之價值。


As high-performance Graphic Processing Unit was widely used on medical devices with rapid replacement of ultrasound machine, automatic measurements such as object detection, image classification and semantic segmentation based on artificial intelligence technology play a significant role on novel medical equipment.

For fetal head measurement, several morphology-based and traditional statistics methods had been used in the past; unfortunately, imprecise prediction result and long executing time usually occurred with these models. Although some deep-learning-based algorithms have been proposed to annotate head circumference for getting occipitofrontal diameter, morphology-based processing should also be utilized to estimate skull thickness for calculating biparietal diameter which is costing large amounts of time while there is no skull annotation dataset provided until the publication of this study.

In this study, we have proposed an end-to-end deep-learning based method named Dual-output Networks designed for annotating head circumference and skull simultaneously. Also, a novel way to generate and process skull annotation from fetal head ultrasound image has been raise. Moreover, a loss function called LOLI Loss based on morphological association of head circumference and skull has been utilized to improve accuracy of prediction. Several commonly used backbone models and attention module have been tried to verify if there is any benefit to improve its precision.

第一章 緒論 9 第二章 相關研究 11 第三章 研究方法 13 3.1 網路架構 13 3.1.1 BackBone骨幹網路 16 3.1.2 單輸出網路 18 3.1.3 雙輸出網路 18 3.2 資料集的生成與製作 20 3.3 Attention注意力機制設計 27 3.4 Loss Function損失函數設計 34 3.4.1 基於像素差異的損失函數 38 3.4.2 基於距離差距的損失函數 38 3.4.3 基於雙輸出相互促進的損失函數(LOLI Loss) 38 第四章 實驗結果 38 4.1 系統架構及硬體規格 38 4.2 實驗資料集及其預處理設計 40 4.3 評量指標 42 4.4 消融實驗 43 4.4.1 不同骨幹網路之效果對比 44 4.4.2 不同Attention機制之效果對比 45 4.4.3 不同LOLI Loss Coefficient之效果對比 47 4.4.4 混合實驗 48 4.5 C++項目的整合、部署與運用實驗 49 第五章 結論 51 5.1 研究成果 51 5.2 未來展望 52 參考資料 54

⦁ Stanislavsky, A., Knipe, H. “Head circumference,” Radiopaedia.org., 2022
⦁ Zador, I.E., Salari, V., Chik, L. and Sokol, R.J., “Ultrasound measurement of the fetal head: computer versus operator,” Ultrasound Obstet Gynecol, 1: 208-211, 2022
⦁ C. W. Hanna, A. B. M. Youssef, “Automated measurements in obstetric ultrasound images,” Proceedings of International Conference on Image Processing, pp. 504-507 vol.3, 1997
⦁ Vikram Chalana, Thomas C. Winter, Dale R. Cyr, David R. Haynor, Yongmin Kim, “Automatic fetal head measurements from sonographic images,” Academic Radiology, Volume 3, Issue 8, Pages 628-635, 1996
⦁ Lu W, Tan J, Floyd R., “Automated fetal head detection and measurement in ultrasound images by iterative randomized Hough transform,” Ultrasound Med Biol, 31(7):929-36., 2005
⦁ Carneiro G, Georgescu B, Good S, Comaniciu D., “Automatic fetal measurements in ultrasound using constrained probabilistic boosting tree,” Med Image Comput Comput Assist Interv., 10(Pt 2):571-9., 2007
⦁ Luo Hong., “An Automatic and Accurate Measurement Method of Fetal Head Circumference Based on Convolutional Neural Network,” CN110279433A., 2019
⦁ Li Zonggui, Zhang Junhua, Mei Liye, “Method for Measuring Fetal Head Circumference in Ultrasound Images Based on Mask R-CNN, ” Chinese Journal of Biomedical Engineering, 40(1): 12-18., 2021
⦁ Mostafa Ghelich Oghli, Ali Shabanzadeh, Shakiba Moradi, Nasim Sirjani, Reza Gerami, Payam Ghaderi, Morteza Sanei Taheri, Isaac Shiri, Hossein Arabi, Habib Zaidi, “Automatic fetal biometry prediction using a novel deep convolutional network architecture,” Physica Medica, Volume 88, Pages 127-137, 2021
⦁ Chen L C, Zhu Y, Papandreou G, et al., “Encoder-decoder with atrous separable convolution for semantic image segmentation,” Proceedings of the European conference on computer vision (ECCV). 801-818., 2018
⦁ Ronneberger, O., Fischer, P., Brox, T., “U-net: Convolutional networks for biomedical image segmentation,” MICCAI., 2015
⦁ Lazebnik, S., Schmid, C., Ponce, J., “Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories,” CVPR, 2006
⦁ Zhao, H., Shi, J., Qi, X., Wang, X., Jia, J., “Pyramid scene parsing network, ” CVPR, 2017
⦁ Xu Cong, Wang Li, “ImageSemantic Segmentation Method Based on Improved DeepLabv3+ Network, Laser & Optoelectronics Progress, 58(16): 1610008., 2021

⦁ Olga Russakovsky*, Jia Deng*, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg and Li Fei-Fei. (* = equal contribution) , “ImageNet Large Scale Visual Recognition Challenge,” IJCV, 2015.
⦁ Sandler M, Howard A, Zhu M, et al., “Mobilenetv2: Inverted residuals and linear bottlenecks,” Proceedings of the IEEE conference on computer vision and pattern recognition. 4510-4520., 2018
⦁ Chollet F., “Xception: Deep learning with depthwise separable convolutions,” Proceedings of the IEEE conference on computer vision and pattern recognition., 1251-1258., 2017
⦁ He K, Zhang X, Ren S, et al., “Deep residual learning for image recognition,” Proceedings of the IEEE conference on computer vision and pattern recognition, 770-778., 2016
⦁ Zhang H, Wu C, Zhang Z, et al., “Resnest: Split-attention networks”, arXiv:2004.08955, 2020
⦁ Frangi A F, Niessen W J, Vincken K L, et al., “Multiscale vessel enhancement filtering,” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer Berlin Heidelberg, 130-137, 1998
⦁ L. M. J. Florack, et al., “Scale and the differential structure of images,”, Imag. and Vis. Comp., 10(6):376–388, 1992.
⦁ R. Liu and D. He, “Semantic Segmentation Based on Deeplabv3+ and Attention Mechanism,” 2021 IEEE 4th Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), pp. 255-259, 2021
⦁ Woo S, Park J, Lee J Y, et al., “Cbam: Convolutional block attention module,” Proceedings of the European conference on computer vision (ECCV). 3-19, 2018
⦁ Fu J, Liu J, Tian H, et al., “Dual attention network for scene segmentation,” Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 3146-3154., 2019
⦁ Liu Y, Shao Z, Hoffmann N., “Global Attention Mechanism: Retain Information to Enhance Channel-Spatial Interactions”, arXiv:2112.05561, 2021
⦁ Liu Y, Shao Z, Teng Y, et al., “NAM: Normalization-based Attention Module,” arXiv:2111.12419, 2021
⦁ Misra D, Nalamada T, Arasanipalai A U, et al., “Rotate to attend: Convolutional triplet attention module,” Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision., 3139-3148., 2021
⦁ Abdollahi A, Pradhan B, Alamri A., “VNet: An end-to-end fully convolutional neural network for road extraction from high-resolution remote sensing data,” IEEE Access, 8: 179424-179436., 2020
⦁ Karimi D, Salcudean S E., “Reducing the hausdorff distance in medical image segmentation with convolutional neural networks,” IEEE Transactions on medical imaging, 39(2): 499-513., 2019
⦁ Thomas L. A. van den Heuvel, Dagmar de Bruijn, Chris L. de Korte and Bram van Ginneken, “Automated measurement of fetal head circumference using 2D ultrasound images,” Zenodo, 2018

QR CODE