簡易檢索 / 詳目顯示

回結果列表
研究生： 游啟芳
You, Qi-Fang
論文名稱： 基於超音波關鍵點評估小兒發展性髖關節發育不良的深度學習方法
Keypoint-based Deep Learning Method in Automatic Evaluation of Developmental Dysplasia of Hip from Pediatric Ultrasound
指導教授： 洪昌鈺
Horng, Ming-Huwi
共同指導教授: 孫永年
Sun, Yung-Nien
學位類別： 碩士
Master
系所名稱： 電機資訊學院 - 資訊工程學系
Department of Computer Science and Information Engineering
論文出版年： 2024
畢業學年度： 112
語文別： 英文
論文頁數： 73
中文關鍵詞： 髖關節發育不良 、髖關節超音波 、關鍵點檢測 、卷積神經網路 、變換器
外文關鍵詞： Developmental Dysplasia of the Hip (DDH), Pediatric Hip Ultrasound, Keypoint Detection, Convolutional Neural Network, Transformer
相關次數： 點閱：58  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 發展性髖關節發育不良(Developmental Dysplasia of the Hip, DDH)是最常見於嬰兒的髖部疾病，因為年紀6個月以下的嬰兒骨頭尚未骨化，無法透過X光確定其位置，因此超音波無侵入性、無輻射的特點，為目前臨床上主要診斷髖關節的方法。
    傳統上，髖關節超音波依賴於骨科醫生的經驗和技術來判斷發展性髖關節發育不良的異常與否，然而，由於影像的成像品質和變異性，診斷過程中可能會出現主觀誤差。這就導致了對於精確、穩定且高效的自動化診斷工具的需求。
    有鑑於卷積神經網路(Convolution Neural Network, CNN)於計算機視覺之出色表現，許多計算機輔助診斷系統(Computer Aided Diagnosis System)從而應運而出，除了降低醫生診斷之時間，亦能發掘可能被遺漏的病徵。本次研究透過基於關鍵點檢測(Keypoint Detection)的深度學習來預測臨床診斷所需之病徵關鍵點，採用卷積神經網路、變換器(Transformer)作為特徵提取的主幹網路，並將注意力模組整合至其中，融合來自主幹網路不同解析度的特徵細節，以挖掘更精細的特徵資訊，達到更精準的關鍵點檢測。
    實驗結果表明通過這些模型網路和模組的結合，採用格雷夫法判斷病症時，達到精確率89.0%，召回率81.0%、準確率90.3%。而採用哈克法判斷病症方面，達到精確率93.4%，召回率88.8%、準確率94.2%。有望提高發展性髖關節發育不良診斷的效率及準確性，從而在醫生臨床診斷的輔助上起到重要作用。

    Developmental dysplasia of the hip (DDH) is a common hip disorder in infants. Since the bones of infants under six months of age have not yet ossified, it is impossible to determine their position through X-rays. Therefore, ultrasound, with its non-invasive and radiation-free characteristics, has become the primary method for diagnosing hip joints in clinical practice.
    Traditionally, hip ultrasound relies on the experience and skill of orthopedic doctors to determine the abnormalities of DDH. However, due to the quality and variability of the imaging, subjective errors can occur during the diagnostic process. This has led to a demand for precise, stable, and efficient automatic diagnostic tools.
    Given the excellent performance of Convolutional Neural Networks (CNN) in computer vision, numerous Computer Aided Diagnosis (CAD) systems have been developed. These systems not only reduce the time cost of doctors' diagnoses but also help uncover potentially overlooked symptoms. In this study, we employ a keypoint detection-based deep learning method, using CNN and Transformer as the backbone networks for feature extraction. We integrate attention modules to combine feature maps from different resolutions of the backbone networks, thereby extracting finer feature information to achieve higher keypoint detection accuracy.
    The experimental results demonstrate that by combining these network models and modules, the precision, recall, and accuracy in diagnosing conditions using the Graf method reached 89.0%, 81.0%, and 90.3%, respectively. In contrast, using the Harcke method for diagnosis, the precision, recall, and accuracy achieved were 93.4%, 88.8%, and 94.2%, respectively. This shows promise in improving the efficiency and accuracy of diagnosing developmental dysplasia of the hip, thus playing a significant role in assisting clinicians with their diagnoses.

    摘要 II ABSTRACT IV 致謝 VI Contents VII List of Tables X List of Figures XIII Chapter 1 INTRODUCTION 1 1.1 Background 1 1.2 Motivation 1 1.3 Neonatal Hip Ultrasound Screening 2 1.4 Related Works 3 1.4.1 Evaluation of DDH Using Deep Learning-Based Methods 3 1.4.2 Keypoint Detection Methods 4 1.4.3 Transfuse 5 1.4.4 Vision Transformer 8 1.4.5 Multi-Scale Feature Representation 11 1.4.6 Feature Fusion Attention Module 12 1.4.7 Attention Gate Module 13 1.4.8 Convolution Block Attention Module 14 1.4.9 Differentiable Spatial to Numerical Transform 16 Chapter 2 METHODOLOGY 18 2.1 Method Overview 18 2.2 Methodof the Automated Hip Dysplasia Diagnosis System 18 2.3 Data Preprocessing and Augmentation 19 2.4 Keypoint Detection Model 19 2.5 Convolutional Neural Network Branch 20 2.6 Transformer Attention Mechanism Branch 23 2.7 Feature Fusion Module 24 2.8 Prediction Head 24 2.9 Loss Function 25 2.10 Keypoint Detection—α and β Angles 27 2.11 Keypoint Detection—Femoral Head Coverage 29 Chapter 3 EXPERIMENTAL RESULTS AND DISSCUSSION 32 3.1 Data collection 32 3.2 Experimental Setup 33 3.3 Five-Fold Cross-Validation 34 3.4 Evaluation Metrics and Experimental Results 35 3.4.1 Keypoint Detection for α Angle and β Angle 35 3.4.2 Keypoint Detection for Femoral Head Coverage 37 3.4.3 Symptom Classification by α Angle and β Angle 39 3.4.4 Symptom Classification by Femoral Head Coverage 41 3.5 Correlation Analysis 42 3.5.1 Pearson correlation coefficient 42 3.5.2 Coefficient of Determination 45 3.6 Statistical Significance 47 3.6.1 Paired Student’s t-Test 47 3.6.2 Mann-Whitney U test 48 3.6.3 Spearman’s Rank Correlations 49 3.7 Ablation Study 50 3.7.1 Ablation Study with Different Convolutional Neural Networks 50 3.7.2 Ablation Study with Different Attention Modules 51 Chapter 4 CONCLUSION AND FUTURE WORK 54 4.1 Conclusion 54 4.2 Future Work 55 Reference 56

    [1] S. o. D. D. o. t. H. Committee on Quality Improvement, "Clinical practice guideline: early detection of developmental dysplasia of the hip," Pediatrics, vol. 105, no. 4, pp. 896-905, 2000.
    [2] O. Furnes, S. Lie, B. Espehaug, S. Vollset, L. Engesaeter, and L. Havelin, "Hip disease and the prognosis of total hip replacements: a review of 53 698 primary total hip replacements reported to the Norwegian arthroplasty register 1987–99," The Journal of Bone & Joint Surgery British Volume, vol. 83, no. 4, pp. 579-579, 2001.
    [3] Y. R. Kang and J. Koo, "Ultrasonography of the pediatric hip and spine," 2017.
    [4] R. Graf, "The diagnosis of congenital hip-joint dislocation by the ultrasonic Combound treatment," Archives of orthopaedic and traumatic surgery, vol. 97, pp. 117-133, 1980.
    [5] H. T. Harcke, N. Clarke, M. Lee, P. F. Borns, and G. D. MacEwen, "Examination of the infant hip with real‐time ultrasonography," Journal of Ultrasound in Medicine, vol. 3, no. 3, pp. 131-137, 1984.
    [6] A. Vaswani et al., "Attention is all you need," Advances in neural information processing systems, vol. 30, 2017.
    [7] Y. Zhang, H. Liu, and Q. Hu, "Transfuse: Fusing transformers and cnns for medical image segmentation," in Medical image computing and computer assisted intervention–MICCAI 2021: 24th international conference, Strasbourg, France, September 27–October 1, 2021, proceedings, Part I 24, 2021: Springer, pp. 14-24.
    [8] M. Zieger, "Ultrasound of the infant hip. Part 2. Validity of the method," Pediatric radiology, vol. 16, no. 6, pp. 488-492, 1986.
    [9] J. Dias, I. Thomas, A. Lamont, and B. Mody, "The reliability of ultrasonographic assessment of neonatal hips," The Journal of Bone & Joint Surgery British Volume, vol. 75, no. 3, pp. 479-482, 1993.
    [10] B. d. C. P. Jacobinoa, M. D. Galvãoa, A. F. da Silvab, and C. C. de Castrob, "Using the Graf method of ultrasound examination to classify hip dysplasia in neonates," Autopsy & case reports 2.2 (2012): 5, 2012.
    [11] J. L. Jaremko, M. Mabee, V. G. Swami, L. Jamieson, K. Chow, and R. B. Thompson, "Potential for change in US diagnosis of hip dysplasia solely caused by changes in probe orientation: patterns of alpha-angle variation revealed by using three-dimensional US," Radiology, vol. 273, no. 3, pp. 870-878, 2014.
    [12] L. Pedrotti, I. Crivellari, A. Degrate, F. De Rosa, F. Ruggiero, and M. Mosconi, "Interpreting neonatal hip sonography: intraobserver and interobserver variability," Journal of Pediatric Orthopaedics B, vol. 29, no. 3, pp. 214-218, 2020.
    [13] H. El-Hariri, K. Mulpuri, A. Hodgson, and R. Garbi, "Comparative evaluation of hand-engineered and deep-learned features for neonatal hip bone segmentation in ultrasound," in Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part II 22, 2019: Springer, pp. 12-20.
    [14] O. Ronneberger, P. Fischer, and T. Brox, "U-net: Convolutional networks for biomedical image segmentation," in Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, 2015: Springer, pp. 234-241.
    [15] A. R. Hareendranathan et al., "Toward automatic diagnosis of hip dysplasia from 2D ultrasound," in 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017), 2017: IEEE, pp. 982-985.
    [16] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," Advances in neural information processing systems, vol. 25, 2012.
    [17] A. Stamper, A. Singh, J. McCouat, and I. Voiculescu, "Infant hip screening using multi-class ultrasound scan segmentation," in 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), 2023: IEEE, pp. 1-4.
    [18] K. He, G. Gkioxari, P. Dollár, and R. Girshick, "Mask r-cnn," in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2961-2969.
    [19] S. Ren, K. He, R. Girshick, and J. Sun, "Faster r-cnn: Towards real-time object detection with region proposal networks," Advances in neural information processing systems, vol. 28, 2015.
    [20] T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature pyramid networks for object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117-2125.
    [21] Z. Cao, G. Hidalgo, T. Simon, S. Wei and Y. Sheikh, "OpenPose: Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields" in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 43, no. 01, pp. 172-186, 2021..
    [22] Y.-P. Chen et al., "Automatic and human level Graf's type identification for detecting developmental dysplasia of the hip," Biomedical Journal, vol. 47, no. 2, p. 100614, 2024.
    [23] Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, "Unet++: A nested u-net architecture for medical image segmentation," in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: 4th International Workshop, DLMIA 2018, and 8th International Workshop, ML-CDS 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 20, 2018, Proceedings 4, 2018: Springer, pp. 3-11.
    [24] S.-W. Lee et al., "Accuracy of new deep learning model-based segmentation and key-point multi-detection method for ultrasonographic developmental dysplasia of the hip (DDH) screening," Diagnostics, vol. 11, no. 7, p. 1174, 2021.
    [25] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770-778.
    [26] A. Dosovitskiy et al., "An image is worth 16x16 words: Transformers for image recognition at scale," arXiv preprint arXiv:2010.11929, 2020.
    [27] A. Newell, K. Yang, and J. Deng, "Stacked Hourglass Networks for Human Pose Estimation," arXiv preprint arXiv:1603.06937, 2016.
    [28] K. Sun, B. Xiao, D. Liu, and J. Wang, "Deep high-resolution representation learning for human pose estimation," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 5693-5703.
    [29] Y. Yuan et al., "Hrformer: High-resolution transformer for dense prediction," arXiv preprint arXiv:2110.09408, 2021.
    [30] J. Schlemper et al., "Attention gated networks: Learning to leverage salient regions in medical images," Medical image analysis, vol. 53, pp. 197-207, 2019.
    [31] S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, "Cbam: Convolutional block attention module," in Proceedings of the European conference on computer vision (ECCV), 2018, pp. 3-19.
    [32] A. Nibali, Z. He, S. Morgan, and L. Prendergast, "Numerical coordinate regression with convolutional neural networks," arXiv preprint arXiv:1801.07372, 2018.

    無法下載圖示 校內：2029-08-29公開
    校外：2029-08-29公開
    電子論文尚未授權公開，紙本請查館藏目錄
    QR CODE