Title Page
Contents
Chapter 1. Introduction 9
Chapter 2. Related Works 12
Chapter 3. Proposed Methods 14
3.1. Overall Framework 14
3.2. Data Preprocessing 16
3.3. Input Patch Generation 17
3.4. Consistency Learning-based Multi-scale Dual-attention Network 19
3.5. Objective Functions 23
Chapter 4. Results 25
4.1. Experimental Datasets 25
4.2. Experimental Settings 27
4.3. Experimental Results 29
Chapter 5. Discussions and Conclusions 37
References 39
국문요약 42
Table 1. Related works of lung tumor segmentation based on deep learning 13
Table 2. Patient characteristics for each dataset used in this study. 25
Table 3. Hyperparameter settings for experiment: The same hyperparameter settings were set except for the loss function of the... 27
Table 4. Patient characteristics for training, validation, and test of each dataset. The amount of data are provided for whole and for each... 28
Table 5. Results of the performance evaluation of lung tumor segmentation: Mean and standard values of the proposed and... 30
Figure 1. Examples of lung tumors of various types, as well as size, location, and shape on chest CT images: (a) small-sized lung... 10
Figure 2. An illustration of overview of the consistency learning-based multi-scale dual-attention segmentation framework. 15
Figure 3. Examples of size-variant patches (first row) and their size-invariant patches (second row): (a) small-sized lung tumor (b)... 18
Figure 4. An illustration of CL-MSDA-Net architecture for lung tumor segmentation. 19
Figure 5. An Illustration of MSDA-Net architecture. 22
Figure 6. An illustration of the proposed module structure: (a) MSDA module (b) Dual-attention module. 22
Figure 7. Results of the qualitative evaluation of lung tumor segmentations according to tumor size group: (a) Original CT image,... 32
Figure 8. Results of the qualitative evaluation of accurately segmented case from 52 test datasets: (a) Original CT image, (b)... 34
Figure 9. Results of the qualitative evaluation of under-segmentation case from 52 test datasets: (a) Original CT image, (b) Ground truth,... 35
Figure 10. Results of the qualitative evaluation of over-segmentation case from 52 test datasets: (a) Original CT image, (b) Ground truth,... 36