成功大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	顏玉晴 Yen, Yu-Qing
論文名稱：	影像區塊導向之骨骼病變檢測方法 A patch-based approach for bone lesion detection
指導教授：	藍崑展 Lan, Kun-chan
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：	2026
畢業學年度：	114
語文別：	英文
論文頁數：	111
中文關鍵詞：	骨骼病變偵測、全身骨掃描、區塊式影像處理、異常偵測、深度學習
外文關鍵詞：	bone lesion detection, bone scintigraphy, patch-based learning, anomaly detection, deep learning
相關次數：	點閱：2 下載：0
分享至:	分享至facebook 分享至twitter

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

本研究提出一套結合結構導向切分與異常感知機制之影像區塊（patch-based）骨骼病變偵測方法，以改善全身骨掃描影像中小病灶不易辨識及影像特性高度不一致所造成的偵測困難。傳統以整張影像為輸入的偵測模型，容易受到全身亮度分布不均與正常高攝取區域的干擾，進而降低對局部骨病灶的敏感度。
雖然 patch-based 學習可提升小病灶可見度，然而現有多數方法採用固定大小切割，容易在切割過程中將病灶截斷或產生僅包含部分病灶訊號的影像區塊，導致監督資訊不足。此外，不同身體部位骨骼結構與攝取特性差異顯著，使得以單一 patch 分佈進行訓練難以有效學習區域專屬病灶特徵。
為解決上述問題，本研究利用關鍵點偵測模型擷取人體解剖結構資訊，並依據身體結構將全身骨掃描影像切分為具語意一致性的 keypoint-based patches，以降低病灶在切割過程中被任意截斷的情形，並促進不同身體區域之病灶特性學習。此外，引入patches，此設計源自骨骼病變常呈現異常放射性攝取的臨床特性。異常偵測模型自動擷取影像中具有非典型亮度分布的區域作為 anomaly-based patches，此設計源自骨骼病變常呈現異常放射性攝取的臨床特性。
本研究於 BS-80K資料庫上進行實驗驗證。結果顯示，所提出之方法可使 Faster R-CNN 基準模型的平均準確率由 59.3% 提升至 68.2%，並在結合 anomaly-based patches 後進一步提升至 71.2%。實驗結果證實，本方法能有效提升全身骨掃描影像中骨骼病變的定位能力。

This study proposes a patch-based framework that incorporates structural awareness and anomaly guidance for bone lesion detection in whole-body bone scintigraphy. Conventional detection models that operate on full resolution images are easily affected by severe intensity heterogeneity and physiologically high uptake regions, which limit their sensitivity to small or subtle bone lesions.
Patch-based learning has been introduced to enhance lesion visibility; however, most existing approaches rely on fixed size patch sampling. Such naive partitioning often fragments lesions or generates patches containing only partial lesion signals, leading to insufficient supervision. Moreover, substantial anatomical and uptake variability across capture lesion characteristics that are specific to individual regions. different body regions makes it difficult for a uniform patch distribution to effectively capture lesion characteristics that are specific to individual regions.
To address these issues, we employ a keypoint detection model to extract anatomical landmarks and partition whole-body bone scans into semantically consistent keypoint-based patches according to human body structure. This structure-informed partitioning reduces arbitrary lesion truncation and enables learning of lesion characteristics in a region specific manner. In addition, an anomaly detection model is incorporated to extract regions with atypical intensity distributions as anomaly-based patches. This design is motivated by the clinical observation that bone lesions often appear as abnormal radiotracer uptake in regions that should not exhibit such intensity patterns.
Experiments conducted on the BS-80K dataset show that the proposed patch-based training strategy with structural awareness and anomaly guidance improves the average precision of a Faster R-CNN baseline from 59.3% to 68.2%, and further increases it to 71.2% when combined with anomaly-based patches. Overall, these results indicate that the proposed framework effectively enhances lesion localization in whole-body bone scintigraphy.

摘要	I
ABSTRACT	III
CONTENTS	V
LIST OF FIGURES	IX
LIST OF TABLES	XI
CHAPTER 1 INTRODUCTION	1
1 WHY AUTOMATED DETECTION OF BONE LESION IS IMPORTANT	1
2 CURRENT CHALLENGES IN BONE LESION DETECTION	1
3 LIMITATION OF EXISTING APPROACHES	5
4 DESIGN MOTIVATION	6
5 OUR APPROACH	7
6 ARCHITECTURE	7
6.1 Offline Phase	8
6.2 Online Phase	10
6.3 Summary of Architectural Advantages	10
CHAPTER 2 RELATED WORK	13
1 PRIOR WORK IN BONE METASTASIS DETECTION	13
1.1 Bone lesion dataset	13
1.2 Bone lesion object detection	14
2 PRIOR WORK IN PATCH-BASE METHOD	16
3 PRIOR WORK IN ANOMALY DETECTION	18
CHAPTER 3 METHOD	21
1 DATA DESCRIPTION	21
1.1 Image Acquisition and Instrumentation	21
1.2 Dataset Statistics and Preprocessing	21
2 KEYPOINT-BASED PATCH GENERATION	22
2.1 Keypoint Detection	22
2.2 Image Preprocessing with CLAHE	24
2.3 Keypoint-based Patch Clipping	25
2.4 Technical Rationale for Patch Strategy and Robustness	26
3 ANOMALY-BASED PATCH GENERATION	27
3.1 Anomaly-Based Patch Selection	28
3.2 Anomaly-Based Patch Cropping	29
4 BASELINE MODEL SELECTION AND REPRODUCIBILITY	30
4.1 Loss Formulations of the Baseline Detector	32
5 PATCH-LEVEL LESION DETECTION	33
5.1 Data Preprocessing	34
5.2 Whole-body Training (Baseline)	35
5.3 Keypoint-based Patch Training	35
5.4 Training with Anomaly-based Patch Augmentation	36
6 INFERENCE PIPELINE FOR WHOLE-BODY BONE LESION DETECTION	37
7 AUXILIARY MODEL PERFORMANCE	38
8 QUANTITATIVE REGIONAL INTENSITY PROFILING METHOD	40
CHAPTER 4 EXPERIMENT RESULTS	43
1 DATA COMPOSITION FOR MODEL TRAINING	43
1.1 Training Data for Keypoint Detection	43
1.2 Training Data for Keypoint-Based Object Detection	43
1.3 Training Data for Anomaly Detection	44
2 EVALUATION METRICS	44
2.1 Precision and Recall	44
2.2 Average Precision (AP)	45
2.3 Anomaly Detection Metrics: AUROC and AUPRO	46
2.4 Keypoint Detection Metric: End-Point Error (EPE)	47
3 ANALYSIS OF BASELINE REPRODUCIBILITY AND RATIONALE FOR PARAMETER OPTIMIZATION	47
4 PERFORMANCE OF TRAINING WITH KEYPOINT-BASED PATCHES	49
5 PERFORMANCE OF TRAINING WITH ANOMALY-BASED PATCHES DATA AUGMENTATION	53
6 VISUALIZATION OF RECONSTRUCTION RESULTS	56
7 INFERENCE EFFICIENCY AND COMPUTATIONAL ENVIRONMENT	57
8 DISCUSSION	59
8.1 Why Training with Keypoint-based Patches Is Effective	59
8.2 Why Training with Anomaly-based Patch Augmentation Is Effective	65
8.3 A High-Uptake Region with Atypical Behavior	67
8.4 Anatomical Ambiguity and Model Uncertainty in Whole-Body Bone Scans	69
8.5 Clinical Evaluation and Region-Specific Threshold Optimization Based on Per-Lesion Clinical Benchmarks	71
8.6 Quantitative Error Analysis and Semantic Ambiguity	74
9 ABLATION STUDY	77
9.1 Ablation Study: Exploratory Hybrid Prediction Strategy	77
9.1.1 Architecture design	77
9.1.2 Performance Analysis	79
9.1.3 Case Discussion: Region-Specific Improvement in left_calf	83
9.2 Ablation Study: Conventional Sliding-Window Patch Cutting	83
9.2.1 Experimental Setup	83
9.2.2 Experimental Observations	84
9.2.3 Analysis: Weak Supervision and Patch Fragmentation	84
9.2.4 Improved Merging Strategy	86
CHAPTER 5 LIMITATIONS AND FUTURE WORK	88
1 LIMITATIONS	88
1.1 Use of a Single Baseline Detection Architecture	88
1.2 Uniform Detector Architecture Across Anatomical Regions	88
1.3 Patch Size Normalization and Background Padding Strategy	89
1.4 Lack of Clinical Context in Image-Only Detection	89
1.5 Limited Exploration of Keypoint and Anomaly Detection Variants	89
2 FUTURE WORK	89
CHAPTER 6 CONCLUSION	91
1 MAIN CONTRIBUTIONS	91
2 EXPERIMENTAL FINDINGS	92
REFERENCE	94
                                    

[1] Charito Love, Anabella S Din, Maria B Tomas, Tomy P Kalapparambath, and Christopher J Palestro, "Radionuclide bone imaging: an illustrative review," Radiographics, vol. 23, no. 2, pp. 341–358, 2003.
[2] Da-Chuan Cheng, Chia-Chuan Liu, Te-Chun Hsieh, Kuo-Yang Yen, and Chia-Hung Kao, "Bone metastasis detection in the chest and pelvis from a whole-body bone scan using deep learning and a small dataset," Electronics, vol. 10, no. 10, p. 1201, 2021.
[3] Pengbo Liu, Hu Han, Yuanqi Du, Heqin Zhu, Yinhao Li, Feng Gu, Honghu Xiao, Jun Li, Chunpeng Zhao, and Li Xiao, "Deep learning to segment pelvic bones: large-scale CT datasets and baseline models," International Journal of Computer Assisted Radiology and Surgery, vol. 16, no. 5, pp. 749–756, 2021.
[4] Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mohammadhadi Bagheri, and R Summers, "Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases," in IEEE CVPR, 2017, vol. 7: sn, p. 46.
[5] Kilian Batzner, Lars Heckler, and Rebecca König, "Efficientad: Accurate visual anomaly detection at millisecond-level latencies," in Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2024, pp. 128–138.
[6] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun, "Faster R-CNN: Towards real-time object detection with region proposal networks," IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137–1149, 2016.
[7] Akinobu Shimizu, Hayato Wakabayashi, Takumi Kanamori, Atsushi Saito, Kazuhiro Nishikawa, Hiromitsu Daisaki, Shigeaki Higashiyama, and Joji Kawabe, "Automated measurement of bone scan index from a whole-body bone scintigram," International journal of computer assisted radiology and surgery, vol. 15, no. 3, pp. 389–400, 2020.
[8] Carolina Elizabeth Villegas-Colmán, Julio César Mello-Román, José Luis Vázquez Noguera, Horacio Legal-Ayala, Pastor Pérez Estigarribia, Benicio Grossling-Vallejos, Ronald Rivas, María Gloria Pedrozo, Cynthia Duarte, and Graciela Giménez, "Bone scan images dataset for study of bone metastases in adult breast cancer patients at IICS-UNA, Paraguay," Data in Brief, vol. 58, p. 111191, 2025.
[9] Zongmo Huang, Xiaorong Pu, Gongshun Tang, Ming Ping, Guo Jiang, Mengjie Wang, Xiaoyu Wei, and Yazhou Ren, "BS-80K: The first large open-access dataset of bone scan images," Computers in Biology and Medicine, vol. 151, p. 106221, 2022.
[10] Chiung-Wei Liao, Te-Chun Hsieh, Yung-Chi Lai, Yu-Ju Hsu, Zong-Kai Hsu, Pak-Ki Chan, and Chia-Hung Kao, "Artificial intelligence of object detection in skeletal scintigraphy for automatic detection and annotation of bone metastases," Diagnostics, vol. 13, no. 4, p. 685, 2023.
[11] Fırat Hardalaç, Fatih Uysal, Ozan Peker, Murat Çiçeklidağ, Tolga Tolunay, Nil Tokgöz, Uğurhan Kutbay, Boran Demirciler, and Fatih Mert, "Fracture detection in wrist X-ray images using deep learning-based object detection models," Sensors, vol. 22, no. 3, p. 1285, 2022.
[12] Le Hou, Dimitris Samaras, Tahsin M Kurc, Yi Gao, James E Davis, and Joel H Saltz, "Patch-based convolutional neural network for whole slide tissue image classification," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2424–2433.
[13] Weimin Chen, Yong Han, Muhammad Awais Ashraf, Junhan Liu, Mu Zhang, Feng Su, Zhiguo Huang, and Kelvin KL Wong, "A patch-based deep learning MRI segmentation model for improving efficiency and clinical examination of the spinal tumor," Journal of Bone Oncology, vol. 49, p. 100649, 2024.
[14] Syed Furqan Qadri, Hongxiang Lin, Linlin Shen, Mubashir Ahmad, Salman Qadri, Salabat Khan, Maqbool Khan, Syeda Shamaila Zareen, Muhammad Azeem Akbar, and Md Belal Bin Heyat, "CT‐based automatic spine segmentation using patch‐based deep learning," International Journal of Intelligent Systems, vol. 2023, no. 1, p. 2345835, 2023.
[15] Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Schölkopf, Thomas Brox, and Peter Gehler, "Towards total recall in industrial anomaly detection," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 14318–14328.
[16] Marco Rudolph, Tom Wehrbein, Bodo Rosenhahn, and Bastian Wandt, "Asymmetric student-teacher networks for industrial anomaly detection," in Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2023, pp. 2592–2602.
[17] Jiawei Yu, Ye Zheng, Xiang Wang, Wei Li, Yushuang Wu, Rui Zhao, and Liwei Wu, "Fastflow: Unsupervised anomaly detection and localization via 2d normalizing flows," arXiv preprint arXiv:2111.07677, 2021.
[18] Zhikang Liu, Yiming Zhou, Yuansheng Xu, and Zilei Wang, "Simplenet: A simple network for image anomaly detection and localization," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 20402–20411.
[19] Stephen M Pizer, "Contrast-limited adaptive histogram equalization: Speed and effectiveness stephen m. pizer, r. eugene johnston, james p. ericksen, bonnie c. yankaskas, keith e. muller medical image display research group," in Proceedings of the first conference on visualization in biomedical computing, Atlanta, Georgia, 1990, vol. 337, p. 2.
[20] Satoshi Suzuki, "Topological structural analysis of digitized binary images by border following," Computer vision, graphics, and image processing, vol. 30, no. 1, pp. 32–46, 1985.
[21] Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár, "Focal loss for dense object detection," in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988.
[22] Alexander Neubeck and Luc Van Gool, "Efficient non-maximum suppression," in 18th international conference on pattern recognition (ICPR'06), 2006, vol. 3: IEEE, pp. 850–855.
[23] Tao Jiang, Peng Lu, Li Zhang, Ningsheng Ma, Rui Han, Chengqi Lyu, Yining Li, and Kai Chen, "Rtmpose: Real-time multi-person pose estimation based on mmpose," arXiv preprint arXiv:2303.07399, 2023.
[24] Kunio Doi, "Computer-aided diagnosis in medical imaging: historical review, current status and future potential," Computerized medical imaging and graphics, vol. 31, no. 4-5, pp. 198–211, 2007.
[25] Maryellen L Giger, Heang‐Ping Chan, and John Boone, "Anniversary paper: history and status of CAD and quantitative image analysis: the role of medical physics and AAPM," Medical physics, vol. 35, no. 12, pp. 5799–5820, 2008.
[26] Hui-Lin Yang, Tao Liu, Xi-Ming Wang, Yong Xu, and Sheng-Ming Deng, "Diagnosis of bone metastases: a meta-analysis comparing 18FDG PET, CT, MRI and bone scintigraphy," European radiology, vol. 21, no. 12, pp. 2604–2617, 2011.
[27] Xiang Yuan, Gong Cheng, Kebing Yan, Qinghua Zeng, and Junwei Han, "Small object detection via coarse-to-fine proposal generation and imitation learning," in Proceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 6317–6327.
[28] Jinwang Wang, Chang Xu, Wen Yang, and Lei Yu, "A normalized Gaussian Wasserstein distance for tiny object detection," arXiv preprint arXiv:2110.13389, 2021.
[29] Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, and Florian Bressand, "Mixtral of experts," arXiv preprint arXiv:2401.04088, 2024.

校外：立即公開

簡易檢索 / 詳目顯示

相關論文