臺灣博碩士論文加值系統

脊椎骨折會造成背痛、脊椎駝背畸形，甚至是嚴重的神經損傷。但在臨床上，診斷脊椎骨折的位置及程度需耗費醫師較多的時間進行量測，因此本研究將深度學習方法應用至脊椎X光影像，輔助醫師從第一手的診斷工具中，進行有效率且準確的診斷。本研究所提出的深度學習方法共分為三個部分：物件檢測(Object detection)、分類(Classification)、語意分割(Segmentation)，透過物件檢測方法偵測出X光影像中各脊椎體的位置，接著針對各脊椎體進行分類及語意分割，透過模型預測該脊椎體是否骨折和其邊緣輪廓，進一步判定骨折位置及骨折程度。研究中使用臺北榮民總醫院所提供的941位病患之脊椎X光影像進行訓練及測試，其腰椎的測試結果表現最佳，物件檢測mAP(mean Average Precision)為90.80%；分類之準確度、靈敏度、特異度分別為92.41%, 91.23%, 93.61%；語意分割之IOU(Interaction over Union)、Dice分別為85.31%, 91.87%。

Vertebral fractures (VFs) are associated with back pain, kyphotic deformity, life-quality disturbance, and increased morbidity and mortality. However on clinical, the orthopedists have to spend much time on measuring the vertebral fracture localization and grading in radiography. Therefore, this study applied the deep learning methods on medical images to assist the orthopedists make more accurate and effective diagnosis. Our model is divided into three parts: object detection, classification and segmentation, used to detect the vertebral body position, discriminate whether it is fracture or not and segment its contour to compute the fracture grade. The test results represent the lumbar spine on radiography performed best. The mAP (mean Average Precision) of object detection was 90.80%. The accuracy, sensitivity and specificity of classification were 92.41%, 91.23% and 93.61%, respectively. The IOU (Interaction over Union) and Dice of segmentation were 85.31% and 91.87%.

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Aims and challenge . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Literature Review 3
2.1 Vertebral Fracture Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.1 Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.2 ResNet and DenseNet . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3.3 fast.ai Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3 Proposed Method 10
3.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Proposed Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.3 Introduction of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4 Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.5 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5.1 Data Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
iv
3.5.2 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5.3 Create Classification Ensemble Model . . . . . . . . . . . . . . . . . . 17
3.6 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6.1 Image Label and Pre-processing . . . . . . . . . . . . . . . . . . . . . 18
3.6.2 U-Net . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Experiment 23
4.1 Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.2 Performance Metrics and Results . . . . . . . . . . . . . . . . . . . . 23
4.2 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.2 Experiment Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.3 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2.4 Experiment Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3.2 Experiment Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3.3 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.4 Experiment Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 Discussion 44
5.1 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6 Conclusion and Future Works 49

[1] ONEILL, T. W., et al. ”Back pain, disability, and radiographic vertebral fracture in European women: a prospective study”. Osteoporosis International, 15.9, pp.760-765, 2004.
[2] COCKERILL, W., et al. ”Health-related quality of life and radiographic vertebral fracture”. Osteoporosis International, 15.2, pp.113-119, 2004.
[3] FINK, Howard A., et al. ”What proportion of incident radiographic vertebral deformities is clinically diagnosed and vice versa?”. Journal of bone and mineral research, 20.7, pp.1216-1222, 2005.
[4] LINDSAY, Robert, et al. ”Risk of new vertebral fracture in the year following a fracture”. Jama, 285.3, pp.320-323, 2001.
[5] LENTLE, Brian C., et al. ”Recognizing and reporting vertebral fractures: reducing the risk of future osteoporotic fractures”. Can Assoc Radiol J, 58.1, pp.27-36, 2007.
[6] FRANCIS, R. M., et al. ”Back pain in osteoporotic vertebral fractures”. Osteoporosis International, 19.7, pp.895-903, 2008.
[7] WONG, Cyrus C.; MCGIRT, Matthew J. ”Vertebral compression fractures: a review of current management and multimodal therapy”. Journal of multidisciplinary healthcare, 6, pp.205, 2013.
[8] DELMAS, Pierre D., et al. ”Underdiagnosis of vertebral fractures is a worldwide problem: the IMPACT study”. Journal of Bone and Mineral Research, 20.4, pp.557-563, 2005.
50
[9] OLCZAK, Jakub, et al. ”Artificial intelligence for analyzing orthopedic trauma radiographs: deep learning algorithmsare they on par with humans for diagnosing fractures?”. Acta orthopaedica, 88.6, pp.581-586, 2017.
[10] BURNS, Joseph E.; YAO, Jianhua; SUMMERS, Ronald M. ”Vertebral body compression fractures and bone density: automated detection and classification on CT images”. Radiology, 284.3, pp.788-797, 2017.
[11] LATEEF, Humaira; PATEL, Deepak. ”What is the role of imaging in acute low back pain?”. Current reviews in musculoskeletal medicine, 2.2, pp.69-73, 2009.
[12] BERRY, Gabriel E., et al. ”Are plain radiographs of the spine necessary during evaluation after blunt trauma? Accuracy of screening torso computed tomography in thoracic/lumbar spine fracture diagnosis”. Journal of Trauma and Acute Care Surgery, 59.6, pp.1410-1413, 2005.
[13] VANDENBERG, James, et al. ”Blunt Thoracolumbar-Spine Trauma Evaluation in the Emergency Department: A Meta-Analysis of Diagnostic Accuracy for History, Physical Examination, and Imaging”. The Journal of emergency medicine, 56.2, pp.153-165, 2019.
[14] GENANT, Harry K., et al. ”Vertebral fracture assessment using a semiquantitative technique”. Journal of bone and mineral research, 8.9, pp.1137-1148, 1993.
[15] REDMON, Joseph, et al. ”You only look once: Unified, real-time object detection”. Proceedings of the IEEE conference on computer vision and pattern recognition, pp.779-788, 2016.
[16] REDMON, Joseph; FARHADI, Ali. ”YOLO9000: better, faster, stronger”. Proceedings of the IEEE conference on computer vision and pattern recognition, pp.7263-7271, 2017.
[17] REDMON, Joseph; FARHADI, Ali. ”Yolov3: An incremental improvement”. arXiv preprint, arXiv:1804.02767, 2018.
[18] PAN, Sinno Jialin; YANG, Qiang. ”A survey on transfer learning”. IEEE Transactions on knowledge and data engineering, 22.10, pp.1345-1359, 2009.
51
[19] RAGHU, Maithra, et al. ”Transfusion: Understanding transfer learning with applications to medical imaging”. arXiv preprint, arXiv:1902.07208, 2019.
[20] HE, Kaiming, et al. ”Deep residual learning for image recognition”. Proceedings of the IEEE conference on computer vision and pattern recognition, pp.770-778, 2016.
[21] HUANG, Gao, et al. ”Densely connected convolutional networks”. Proceedings of the IEEE conference on computer vision and pattern recognition, pp 4700-4708, 2017.
[22] SMITH, Leslie N. ”Cyclical learning rates for training neural networks”. IEEE Winter Conference on Applications of Computer Vision (WACV), pp. 464-472, 2017.
[23] SMITH, Leslie N. ”A disciplined approach to neural network hyper-parameters: Part 1–learning rate, batch size, momentum, and weight decay”. arXiv preprint, arXiv:1803.09820, 2018.
[24] LONG, Jonathan; SHELHAMER, Evan; DARRELL, Trevor. ”Fully convolutional networks for semantic segmentation”. Proceedings of the IEEE conference on computer vision and pattern recognition., pp.3431-3440, 2015..
[25] RONNEBERGER, Olaf; FISCHER, Philipp; BROX, Thomas. ”Unet: Convolutional networks for biomedical image segmentation”. International Conference on Medical image computing and computer-assisted intervention, Springer, Cham, pp.234-241, 2015.
[26] RAJPURKAR, Pranav, et al. ”Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning”. arXiv preprint, arXiv:1711.05225, 2017.
[27] SELVARAJU, Ramprasaath R., et al. ”Grad-cam: Visual explanations from deep networks via gradient-based localization”. Proceedings of the IEEE International Conference on Computer Vision, pp.618-626, 2017.
52
[28] BURNS, Joseph E.; YAO, Jianhua; SUMMERS, Ronald M. ”Vertebral body compression fractures and bone density: automated detection and classification on CT images.” Radiology, 284.3, pp.788-797, 2017.
[29] BINKLEY, Neil, et al. ”Lateral vertebral assessment: a valuable technique to detect clinically significant vertebral fractures”. Osteoporosis international, 16.12, pp.1513-1518, 2005