舟狀骨骨折是十分常見的手腕傷害，醫生通常會通過拍攝X光圖進行診斷治療，這個過程通常需要花費許多時間。在多種舟骨骨折類型中，特別是隱匿性骨折和非移位性骨折，由於其外觀微妙且骨密度變化性強，使用X光圖方法診斷難度極大，造成診斷上的困難。鑒於去噪擴散模型在關於醫學影像處理任務上的良好表現，不但能大幅降低醫生診斷的時間，也能精準定位可能的骨折區域，找出被肉眼忽視的病症。

本研究提出了一個基於擴散模型的舟狀骨骨折偵測與定位框架。整體架構的第一階段，我們通過嵌入骨折異常的方式擴充訓練數據集。我們在健康舟狀骨圖像上生成偽斷裂骨折區域，輸出健康以及骨折兩份資料集，這一階段形成了自監督的學習策略，避免了複雜且耗時的人工醫學影像標註過程。第二階段，基於擴散模型的重建模型學習健康舟狀骨影像的特徵，對其他偽骨折斷裂舟骨資料集中的影像進行高質量重建，得到健康狀態的舟骨影像。第三階段通過類 U-Net 網絡識別目標影像與重建影像之間的差異，該差異作為目標舟骨影像是否骨折的判斷依據。

模型訓練完畢之後，我們可以通過模型對真實的舟骨影像進行骨折的診斷，我們將模型診斷結果與醫生進一步標註的骨折精確位置共同進行模型效能的評估。我們的方法在舟骨斷裂偵測效果上用Image AUROC進行評估，取得的成績是0.993。也就是說，我們的模型在舟骨骨折分類的評估上，正確率達到了0.983，召回率達到了1.00，準確率達到了0.975。我們在舟骨骨折定位上用Pixel AUROC 、Pixel Region Overlap進行評估，取得的成績分別是0.978，0.921。從實驗中看出，通過擴散模型進行舟骨骨折的診斷，模型能夠精準診斷出該病人是否骨折，並且能夠精準定位出骨折位置。

Scaphoid fractures are a common wrist injury, typically diagnosed and treated through X-ray imaging, a process that is often time-consuming. Among the various types of scaphoid fractures, occult and non-displaced fractures pose significant diagnostic challenges due to their subtle appearance and variable bone density, complicating accurate identification via X-ray images. Given the strong performance of denoising diffusion models in medical image processing tasks, these models can significantly reduce diagnostic time while precisely localizing potential fracture regions, identifying conditions overlooked by the human eye.

This study proposes a scaphoid fracture detection and localization framework based on diffusion models. In the first stage, we augment the training dataset by embedding fracture anomalies. Pseudo-fracture regions are generated on healthy scaphoid images, producing both healthy and fractured datasets, forming a self-supervised learning strategy that avoids the complex and time-consuming manual annotation of medical images. In the second stage, a diffusion-based reconstruction model learns the features of healthy scaphoid images to perform high-quality reconstruction of pseudo-fractured scaphoid images, generating healthy scaphoid images. In the third stage, a U-Net-like network identifies differences between the target and reconstructed images, using these differences to determine whether the target scaphoid image contains a fracture.

After model training, we evaluated its diagnostic performance on real scaphoid images by comparing the model’s results with precise fracture locations further annotated by physicians. The proposed method achieved an Image AUROC of 0.993 for scaphoid fracture detection, corresponding to an accuracy of 0.983, a recall of 1.00, and a precision of 0.975. For fracture localization, the model achieved a Pixel AUROC of 0.978 and a Pixel Region Overlap (PRO) of 0.921. Experimental results demonstrate that the diffusion-based model can accurately diagnose scaphoid fractures and precisely localize fracture regions.

[1]Hasan Iqbal, Umar Khalid, Chen Chen, and Jing Hua. Unsupervised anomaly detection in medical images using masked diffusion model. In International Work￾shop on Machine Learning in Medical Imaging, pages 372–381. Springer, 2023.
[2]Hui Zhang, Zheng Wang, Dan Zeng, Zuxuan Wu, and Yu-Gang Jiang. Diffu￾sionad: Norm-guided one-step denoising diffusion for anomaly detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025.
[3]Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Mu￾nich, Germany, October 5-9, 2015, proceedings, part III 18, pages 234–241. Springer, 2015.
[4]Jieneng Chen, Yongyi Lu, Qihang Yu, Xiangde Luo, Ehsan Adeli, Yan Wang, Le Lu, Alan L Yuille, and Yuyin Zhou. Transunet: Transformers make strong encoders for medical image segmentation. arXiv preprint arXiv:2102.04306, 2021.
[5]Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132– 7141, 2018.
[6]Nils Hendrix. Artificial intelligence for computer aided diagnosis of scaphoid frac￾tures and associated instability on conventional radiography. 2024.
[7]Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020.
[8]Julia Wolleb, Florentin Bieder, Robin Sandk¨uhler, and Philippe C Cattin. Diffu￾sion models for medical anomaly detection. In International Conference on Medical image computing and computer-assisted intervention, pages 35–45. Springer, 2022.
[9]Roberto Di Via, Francesca Odone, and Vito Paolo Pastore. Self-supervised pre-training with diffusion model for few-shot landmark detection in x-ray im￾ages. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3886–3896. IEEE, 2025.
[10]Alfred P Yoon, Yi-Lun Lee, Robert L Kane, Chang-Fu Kuo, Chihung Lin, and Kevin C Chung. Development and validation of a deep learning model using convolutional neural networks to identify scaphoid fractures in radiographs. JAMA network open, 4(5):e216096–e216096, 2021.
[11]Nils Hendrix, Ernst Scholten, Bastiaan Vernhout, Stefan Bruijnen, Bas Maresch, Mathijn de Jong, Suzanne Diepstraten, Stijn Bollen, Steven Schalekamp, Maarten de Rooij, et al. Development and validation of a convolutional neural network for automated detection of scaphoid fractures on conventional radiographs. Radiology: Artificial Intelligence, 3(4):e200260, 2021.
[12]Nils Hendrix, Ward Hendrix, Kees van Dijke, Bas Maresch, Mario Maas, Stijn Bollen, Alexander Scholtens, Milko de Jonge, Lee-Ling Sharon Ong, Bram van Ginneken, et al. Musculoskeletal radiologist-level performance by using deep learn￾ing for detection of scaphoid fractures on conventional multi-view radiographs of hand and wrist. European Radiology, 33(3):1575–1588, 2023.
[13]Amanpreet Singh, Ali Abbasian Ardakani, Hui Wen Loh, PV Anamika, U Ra￾jendra Acharya, Sidharth Kamath, and Anil K Bhat. Automated detection of scaphoid fractures using deep neural networks in radiographs. Engineering Appli￾cations of Artificial Intelligence, 122:106165, 2023.
[14]Matan Kraus, Roi Anteby, Eli Konen, Iris Eshed, and Eyal Klang. Artificial intel￾ligence for x-ray scaphoid fracture detection: a systematic review and diagnostic test accuracy meta-analysis. European radiology, 34(7):4341–4351, 2024.
[15]Cheng Chen, Bo Liu, Kangneng Zhou, Wanzhang He, Fei Yan, Zhiliang Wang, and Ruoxiu Xiao. Csr-net: Cross-scale residual network for multi-objective scaphoid fracture segmentation. Computers in Biology and Medicine, 137:104776, 2021.
[16]A B¨utzow, TT Anttila, V Haapam¨aki, and J Ryh¨anen. A novel segmentation￾based deep learning model for enhanced scaphoid fracture detection. European Journal of Radiology, page 112309, 2025.
[17]李榮祥. 舟狀骨骨折檢測與多角度影像分類之深度學習算法. Master’s thesis, 國立成功大學, 2022.
[18]Yijun Yang, Huazhu Fu, Angelica I Aviles-Rivero, Carola-Bibiane Sch¨onlieb, and Lei Zhu. Diffmic: Dual-guidance diffusion network for medical image classification. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 95–105. Springer, 2023.
[19]Tianyu Lin, Zhiguang Chen, Zhonghao Yan, Weijiang Yu, and Fudan Zheng. Sta￾ble diffusion segmentation for biomedical images with single-step reverse process. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 656–666. Springer, 2024.
[20]Tim Salimans, Andrej Karpathy, Xi Chen, and Diederik P Kingma. Pixelcnn++: Improving the pixelcnn with discretized logistic mixture likelihood and other mod￾ifications. In International Conference on Learning Representations, 2017.
[21]Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.
[22]Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017.
[23]Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34:8780–8794, 2021.
[24]Xihui Liu, Dong Huk Park, Samaneh Azadi, Gong Zhang, Arman Chopikyan, Yuxiao Hu, Humphrey Shi, Anna Rohrbach, and Trevor Darrell. More control for free! image synthesis with semantic diffusion guidance. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 289–299, 2023.
[25]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, 39(6):1137–1149, 2016.
[26]Tsung-Yi Lin, Piotr Doll´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2117–2125, 2017.
[27]Trevor J Chan and Chamith S Rajapakse. A super-resolution diffusion model for recovering bone microstructure from ct images. Radiology: Artificial Intelligence, 5(6):e220251, 2023.
[28]Hongxu Jiang, Muhammad Imran, Teng Zhang, Yuyin Zhou, Muxuan Liang, Kuang Gong, and Wei Shao. Fast-ddpm: Fast denoising diffusion probabilistic models for medical image-to-image generation. IEEE Journal of Biomedical and Health Informatics, 2025.
[29]Victor Livernoche, Vineet Jain, Yashar Hezaveh, and Siamak Ravanbakhsh. On diffusion modeling for anomaly detection. In The Twelfth International Conference on Learning Representations, 2024.
[30]Teng Hu, Jiangning Zhang, Ran Yi, Yuzhen Du, Xu Chen, Liang Liu, Yabiao Wang, and Chengjie Wang. Anomalydiffusion: Few-shot anomaly image generation with diffusion model. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 8526–8534, 2024.
[31]Tai-Hua Yang, Ming-Huwi Horng, Rong-Shiang Li, and Yung-Nien Sun. Scaphoid fracture detection by using convolutional neural network. Diagnostics, 12(4):895, 2022.
[32]Tai-Hua Yang, Yung-Nien Sun, Rong-Shiang Li, and Ming-Huwi Horng. The detection and classification of scaphoid fractures in radiograph by using a convolutional neural network. Diagnostics, 14(21):2425, 2024.