臺灣博碩士論文加值系統

本研究旨在建立一套支氣管內視鏡超聲波肺部病灶的電腦輔助診斷系統協助醫師找出病灶區，在EBUS-TBNA手術過程中，醫師可以利用灰階超聲影像判斷出病灶位置，然而這些影像中包含大量雜訊且會受其他組織或是血管影響，大多難以判讀。先前研究缺乏目標檢測模型在EBUS-TBNA的應用，且尚無對於EBUS-TBNA資料集中標註資料少提出良好的解決方法，而其他超聲影像上的目標檢測模型相關研究，雖都能在各自任務上捕捉到目標區域，但由於訓練以及預測皆是用二維圖像，因此沒辦法透過捕捉時間維度的特徵去改善預測值。本研究提出了一套基於三維影像的目標檢測模型，首先會透過生成式模型生成一組較好的查詢，之後藉由注意力機制捕捉時間上的關聯性，並在後續利用篩選機制挑選出適合的先前幀資訊傳遞到當前幀，之後用老師學生模型訓練方法進一步利用未標註資料優化模型，透過加入不同的資料增強以及特徵對齊，使模型可以具備一定的抗干擾能力。測試結果顯示，此模型使用捕捉時空資訊以及半監督學習的方法，在測試資料集上取得了48.7 AP，相較其他模型表現得更好，且也達到了79.2 AR大幅領先了其他現有模型，大幅降低了病灶漏檢率。

This study aims to establish a computer-aided diagnostic system for lung lesions using endobronchial ultrasound (EBUS) to assist physicians in identifying lesion areas. During EBUS-transbronchial needle aspiration (EBUS-TBNA) procedures, physicians rely on grayscale ultrasound images to determine the location of lesions. However, these images often contain significant noise and can be influenced by surrounding tissues or blood vessels, making interpretation challenging. Previous research has lacked the application of object detection models to EBUS-TBNA, and there has been no well-defined solution for the lack of annotated data in the EBUS-TBNA dataset. In related studies on ultrasound images, although models have been successful in capturing target regions for their respective tasks, their training and predictions have been based on two-dimensional images, limiting their ability to leverage temporal features for improved predictions. This study introduces a three-dimensional image-based object detection model. It first generates a set of improved queries using a diffusion model, then captures temporal correlations through an attention mechanism. A filtering mechanism selects relevant information from previous frames to pass to the current frame. Subsequently, a teacher-student model training approach is employed to further optimize the model using unlabeled data. By incorporating various data augmentation and feature alignment, the model gains robustness against interference. Test results demonstrate that this model, which captures spatiotemporal information and employs semi-supervised learning methods, achieves an Average Precision (AP) of 48.7 on the test dataset, outperforming other models. It also achieves an Average Recall (AR) of 79.2, significantly reducing the miss rate of lesions and leading over existing models.

中文摘要i
Abstractii
致謝iv
目錄v
圖目錄vii
表目錄viii
第一章緒論1
1.1研究背景1
1.2文獻回顧2
1.3研究動機與目標5
1.4論文架構7
第二章實驗原理8
2.1DEtection TRansformer (DETR)8
2.2半監督學習方法9
2.3Diffusion model10
第三章實驗方法12
3.1實驗說明及進行步驟12
3.1.1 實驗概述12
3.1.2 實驗流程12
3.1.3 EBUS-TBNA資料來源13
3.2前處理和資料增強14
3.3模型架構15
3.3.1 SVDETR (Short time video detection transformer)15
3.3.2 生成式擴散模型 (Diffusion model)17
3.3.3 半監督學習(Teacher student model)17
3.4超參數設定18
3.5實驗設備19
第四章實驗結果與討論20
4.1評價指標20
4.2各模型的比較21
4.3隨機遮罩數量對於模型性能的影響23
4.4不同擴散模型策略應用於模型的結果比較23
4.5不同數量篩選機制結果比較24
4.6半監督學習應用在影像物體檢測模型的比較25
4.7模型運行速度測試26
4.8實驗總結27
第五章結論與未來研究內容28
參考文獻29

[ 1 ]Jemal A, Siegel R, Ward E, Hao Y, Xu J, Thun MJ. Cancer statistics, 2009. CA Cancer J Clin 2009, 59:225-49.
[ 2 ]Yasufuku K, Fujisawa T. Staging and diagnosis of non‐small cell lung cancer: Invasive modalities. Respirology 2007, 12:173-83.
[ 3 ]Lin CK, Lai CL, Chang LY, Wen YF, Ho CC. Learning curve and advantages of endobronchial ultrasound‐guided transbronchial needle aspiration as a first‐line diagnostic and staging procedure. Thorac Cancer 2019, 75-82.
[ 4 ] Nakajima T, Yasufuku K, Kurosu K, Takiguchi Y, Fujiwara T, Chiyo M, et al. The role of EBUS-TBNA for the diagnosis of sarcoidosis – comparisons with other bronchoscopic diagnostic modalities. Respir Med 2009, 103:1796-800.
[ 5 ]Navani N, Nankivell M, Lawrence DR, Lock S, Makker H, Baldwin DR, et al. Lung cancer diagnosis and staging with endobronchial ultrasound-guided transbronchial needle aspiration compared with conventional approaches: an open-label, pragmatic, randomised controlled trial. Lancet Respir Med 2015, 3:282–9.
[ 6 ]Yasufuku K, Chiyo M, Sekine Y, Chhajed PN, Shibuya K, Iizasa T, et al. Real-time Endobronchial Ultrasound-Guided Transbronchial Needle Aspiration of Mediastinal and Hilar Lymph Nodes. Chest 2004, 126:122–8.
[ 7 ]Fujiwara T, Yasufuku K, Nakajima T, Chiyo M, Yoshida S, Suzuki M, et al. The Utility of Sonographic Features During Endobronchial Ultrasound-Guided Transbronchial Needle Aspiration for Lymph Node Staging in Patients With Lung Cancer. Chest 2010, 138:641–7.
[ 8 ]Nakajima T, Anayama T, Shingyoji M, Kimura H, Yoshino I, Yasufuku K. Vascular Image Patterns of Lymph Nodes for the Prediction of Metastatic Disease During EBUS-TBNA for Mediastinal Staging of Lung Cancer. J Thorac Oncol 2012, 7:1009–14.
[ 9 ]Izumo T, Sasada S, Chavez C, Matsumoto Y, Tsuchida T. Endobronchial Ultrasound Elastography in the Diagnosis of Mediastinal and Hilar Lymph Nodes. Jpn J Clin Oncol 2014, 44:956–62.
[ 10 ]LIN, Ching-Kai, et al. TransEBUS: The interpretation of endobronchial ultrasound image using hybrid transformer for differentiating malignant and benign mediastinal lesions. Journal of the Formosan Medical Association, 2024.
[ 11 ]YONG, Seung Hyun, et al. Malignant thoracic lymph node classification with deep convolutional neural networks on real-time endobronchial ultrasound (EBUS) images. Translational Lung Cancer Research, 2022, 11.1: 14.
[ 12 ]ITO, Yuki, et al. Prediction of nodal metastasis in lung cancer using deep learning of endobronchial ultrasound images. Cancers, 2022, 14.14: 3334
[ 13 ]ZENG, Xianhua, et al. Deep learning for ultrasound image caption generation based on object detection. Neurocomputing, 2020, 392: 132-141.
[ 14 ]BOSE, Anu, et al. Faster RCNN hyperparameter selection for breast lesion detection in 2D ultrasound images. Springer International Publishing, 2022, 179-190.
[ 15 ]SNIDER, Eric J., et al. Evaluation of an object detection algorithm for shrapnel and development of a triage tool to determine injury severity. Journal of imaging, 2022, 8.9: 252.
[ 16 ]CAO, Guimei, et al. Feature-fused SSD: Fast detection for small objects. In: Ninth international conference on graphic and image processing (ICGIP 2017). SPIE, 2018, 381-388.
[ 17 ]TANG, Yiwen, et al. Detection of spine curve and vertebral level on ultrasound images using detr. In: 2022 IEEE International Ultrasonics Symposium (IUS). IEEE, 2022, 1-4.
[ 18 ]FUJITAKE, Masato; SUGIMOTO, Akihiro. Video sparse transformer with attention-guided memory for video object detection. IEEE Access, 2022, 10: 65886-65900.
[ 19 ]ZHOU, Qianyu, et al. TransVOD: end-to-end video object detection with spatial-temporal transformers. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022.
[ 20 ]WANG, Han, et al. Ptseformer: Progressive temporal-spatial enhanced transformer towards video object detection. In: European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022, 732-747.
[ 21 ]ROH, Si-Dong; CHUNG, Ki-Seok. DiffusionVID: Denoising Object Boxes with Spatio-temporal Conditioning for Video Object Detection. IEEE Access, 2023.
[ 22 ]ZHANG, Jiacheng, et al. Semi-detr: Semi-supervised object detection with detection transformers. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023, 23809-23818.
[ 23 ]WANG, Xinjiang, et al. Consistent targets provide better supervision in semi-supervised object detection. 2022.
[ 24 ]LIU, Yen-Cheng; MA, Chih-Yao; KIRA, Zsolt. Unbiased teacher v2: Semi-supervised object detection for anchor-free and anchor-based detectors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022, 9819-9828.
[ 25 ]ZHU, Xizhou, et al. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010, 04159, 2020.
[ 26 ]DAI, Xiyang, et al. Dynamic detr: End-to-end object detection with dynamic attention. In: Proceedings of the IEEE/CVF international conference on computer vision. 2021, 2988-2997.
[ 27 ]CARION, Nicolas, et al. End-to-end object detection with transformers. In: European conference on computer vision. Cham: Springer International Publishing, 2020, 213-229.
[ 28 ]TARVAINEN, Antti; VALPOLA, Harri. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems, 2017, 30.
[ 29 ]CAI, Zhaowei, et al. Semi-supervised vision transformers at scale. Advances in Neural Information Processing Systems, 2022, 35: 25697-25710.
[ 30 ]HO, Jonathan; JAIN, Ajay; ABBEEL, Pieter. Denoising diffusion probabilistic models. Advances in neural information processing systems, 2020, 33: 6840-6851.
[ 31 ]NICHOL, Alexander Quinn; DHARIWAL, Prafulla. Improved denoising diffusion probabilistic models. In: International conference on machine learning. PMLR, 2021, 8162-8171.
[ 32 ]CHEN, Shoufa, et al. Diffusiondet: Diffusion model for object detection. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, 19830-19843.
[ 33 ]Ge, Z., Liu, S., Wang, F., Li, Z., & Sun, J. (2021). Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430.
[ 34 ]REN, Shaoqing, et al. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE transactions on pattern analysis and machine intelligence, 2016, 39.6: 1137-1149.
[ 35 ]Shi, J., & Xu, C. Differential Network for Video Object Detection.
[ 36 ]ZHU, Xizhou, et al. Flow-guided feature aggregation for video object detection. In: Proceedings of the IEEE international conference on computer vision. 2017, 408-417.
[ 37 ]DENG, Jiajun, et al. Relation distillation networks for video object detection. In: Proceedings of the IEEE/CVF international conference on computer vision. 2019, 7023-7032.
[ 38 ]CHEN, Yihong, et al. Memory enhanced global-local aggregation for video object detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020, 10337-10346.
[ 39 ]ROH, Si-Dong; CHUNG, Ki-Seok. DAFA: Diversity-aware feature aggregation for attention-based video object detection. IEEE Access, 2022, 10: 93453-93463.
[ 40 ]SHI, Yuheng; WANG, Naiyan; GUO, Xiaojie. YOLOV: Making still image object detectors great at video object detection. In: Proceedings of the AAAI Conference on Artificial Intelligence. 2023, 2254-2262.