近年來，有越來越多的醫學影像問題是透過深度學習的方法來解決，不僅可以減輕醫師的負擔，還可以提供較一致的結果。在本文中，我們將大腸息肉分割模型 HarDNet-MSEG 進行改良，並結合一些新方法，將其應用在電腦斷層資料的肝臟腫瘤分割任務。HarDNet-MSEG 使用的骨幹網路是 HarDNet，這是一個擁有較少參數和計算量的高效能模型。其增強模組使用的是 Receptive Field Block，通過使用不同大小的空洞卷積來學習多尺度的特徵。解碼器則是使用 Dense Aggregation 的方法來融合不同層級的特徵，可以減少信息損失並提高分割精度。我們提出的 HarDNet-LiTS 模型是在 HarDNet-MSEG 的增強模組中加入注意力機制，並使用相鄰切片取代單一切片來作為輸入資料，在訓練過程中使用了彈性變換來增加輸入資料的多樣性，讓模型可以學習到更多不同的資訊，以提升分割的性能。在推理時，我們使用了 Test Time Augmentation 方法，讓模型可以擁有精準的分割結果。最後，我們的方法在 LiTS 公開數據集上達到了 0.7886 的 slice DSC以及 0.7006 的 volume DSC 成績。

In recent years, an increasing number of medical imaging problems have been solved through deep learning methods. These methods not only reduce the burden on doctors but also provide more consistent results. In this thesis, we improved the polyp segmentation model HarDNet-MSEG and incorporated some new methods to apply it to the liver tumor segmentation tasks in computed tomography (CT) scan data. HarDNet-MSEG utilizes HarDNet as its backbone, which is an efficient model with fewer parameters and lower computational requirements. Its enhancement module employs Receptive Field Blocks to learn multi-scale features by utilizing different sizes of dilated convolutions. The decoder uses a Dense Aggregation method to integrate features from different levels, reducing information loss and improving segmentation accuracy. Our proposed HarDNet-LiTS model adds an attention mechanism to the enhancement module of HarDNet-MSEG. In addition, we use adjacent slices instead of a single slice as input data and apply elastic transformations during training to increase input data diversity, allowing the model to learn more varied information to improve segmentation performance. During inference, we use Test Time Augmentation (TTA) to achieve more precise segmentation results. Our method achieves a slice DSC of 0.7886 and a volume DSC of 0.7006 on the LiTS public dataset.

Acknowledgements
摘要 i
Abstract ii
1 Introduction 1
2 Related Works 5
2.1 Medical Segmentation Architectures 5
2.1.1 Two-dimensional Networks 5
2.1.2 Three-dimensional Networks 6
2.2 Attention Mechanisms 6
3 Proposed Methods 9
3.1 Overall Segmentation Process 9
3.2 Encoder 10
3.3 Enhancement Modules 10
3.3.1 Receptive Field Block 11
3.3.2 Attention Mechanisms 11
3.4 Decoder 12
3.5 HarDNet-SE-RFB 13
3.6 Loss Function 13
4 Experiments 15
4.1 Dataset 15
4.2 Training Settings 15
4.2.1 Pre-Processing 16
4.2.2 Post-Processing 16
4.3 Metrics 16
4.4 Liver Segmentation 17
4.5 Ablation Study of Tumor Segmentation 18
4.5.1 Comparison of Different Attention Mechanisms 18
4.5.2 Comparison of Different Models 19
4.5.3 Effectiveness of Adjacent Slices Method 20
4.5.4 Effectiveness of Elastic Transformation 21
4.5.5 Comparison of Test Time Augmentation Methods 23
4.6 Comparison 24
5 Conclusions and Future Work 27
5.1 Conclusions 27
5.2 Future Work 28
References 29
Appendix 31

[1] P. Bilic, P. Christ, H. B. Li, E. Vorontsov, A. Ben-Cohen, G. Kaissis, A. Szeskin, C. Jacobs, G. E. H. Mamani, G. Chartrand, et al., "The liver tumor segmentation benchmark (lits),”Medical Image Analysis, vol. 84, p. 102680, 2023.
[2] 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, pp. 234–241, Springer, 2015.
[3] 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, pp. 3–11, Springer, 2018.
[4] O. Oktay, J. Schlemper, L. L. Folgoc, M. Lee, M. Heinrich, K. Misawa, K. Mori, S. McDonagh, N. Y. Hammerla, B. Kainz, et al., “Attention u-net: Learning where to look for the pancreas,” arXiv preprint arXiv:1804.03999, 2018.
[5] F. I. Diakogiannis, F. Waldner, P. Caccetta, and C. Wu, “Resunet-a: A deep learning framework for semantic segmentation of remotely sensed data,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 162, pp. 94–114, 2020.
[6] C.-H. Huang, H.-Y. Wu, and Y.-L. Lin, “Hardnet-mseg: A simple encoder-decoder polyp segmentation neural network that achieves over 0.9 mean dice and 86 fps,” arXiv preprint arXiv:2101.07172, 2021.
[7] S. Targ, D. Almeida, and K. Lyman, “Resnet in resnet: Generalizing residual architectures,” arXiv preprint arXiv:1603.08029, 2016.
[8] D. Jha, P. H. Smedsrud, M. A. Riegler, D. Johansen, T. De Lange, P. Halvorsen, and H. D. Johansen, “Resunet++: An advanced architecture for medical image segmentation,” in 2019 IEEE international symposium on multimedia (ISM), pp. 225–2255, IEEE, 2019.
[9] T. Fan, G. Wang, Y. Li, and H. Wang, “Ma-net: A multi-scale attention network for liver and tumor segmentation,” IEEE Access, vol. 8, pp. 179656–179665, 2020.
[10] Y. Wang, T. Wang, H. Li, and H. Wang, “Acf-transunet: Attention-based coarse-fine transformer u-net for automatic liver tumor segmentation in ct images,” in 2023 4th International Conference on Big Data & Artificial Intelligence & Software Engineering (ICBASE), pp. 84–88, IEEE, 2023.
[11] J. Chen, Y. Lu, Q. Yu, X. Luo, E. Adeli, Y. Wang, L. Lu, A. L. Yuille, and Y. Zhou,“Transunet: Transformers make strong encoders for medical image segmentation,” arXiv preprint arXiv:2102.04306, 2021.
[12] Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, and O. Ronneberger, “3d u-net: learning dense volumetric segmentation from sparse annotation,” in Medical Image Computing and Computer-Assisted Intervention–MICCAI 2016: 19th International Conference, Athens, Greece, October 17-21, 2016, Proceedings, Part II 19, pp. 424–432, Springer, 2016.
[13] J. M. J. Valanarasu, V. A. Sindagi, I. Hacihaliloglu, and V. M. Patel, “Kiu-net: Towards accurate segmentation of biomedical images using over-complete representations,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part IV 23, pp. 363–373, Springer, 2020.
[14] P. Chao, C.-Y. Kao, Y.-S. Ruan, C.-H. Huang, and Y.-L. Lin, “Hardnet: A low memory traffic network,” in Proceedings of the IEEE/CVF international conference on computer vision, pp. 3552–3561, 2019.
[15] S. Liu, D. Huang, et al., “Receptive field block net for accurate and fast object detection,”in Proceedings of the European conference on computer vision (ECCV), pp. 385–400, 2018.
[16] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 7132–7141, 2018.
[17] 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), pp. 3–19, 2018.
[18] M. Antonelli, A. Reinke, S. Bakas, K. Farahani, A. Kopp-Schneider, B. A. Landman, G. Litjens, B. Menze, O. Ronneberger, R. M. Summers, et al., “The medical segmentation decathlon,” Nature communications, vol. 13, no. 1, p. 4128, 2022.