Title Page
Contents
Summary 6
1. Introduction 9
2. Methods 11
1. Data 11
1) Private Datasets 11
2) Public datasets 12
2. Vessel extraction 12
1) Network architecture 13
2) Learning parameters 14
3) Loss function 14
3. Aneurysm detection 14
1) Surface mesh parcellation 15
2) Point cloud representation learning for classification 15
3) Prediction aggregation 16
4) Implementation 16
4. Experiments 16
5. Evaluation metrics 17
3. Results 19
1. Cross-modality performance 19
2. Multi-site performance 20
1) Performance on healthy subjects 21
3. Zi-Hao Study repetition 22
4. Discussion 24
5. Conclusion 27
6. References 28
국문요약 30
〈Table 1〉 The performance results of previously reported machine learning models for detecting intracranial aneurysms 10
〈Table 2〉 The results of the Cross-Modality Performance experiments presented in relation to the size of the IA. 20
〈Table 3〉 Results of Multi-site performance experiments. 22
〈Table 4〉 Zi-Hao Bo Comparison experiment results are presented in relation to the size of the IA. 23
[Figure 1] Network Architecture as generated by nnUNet 13
[Figure 2] Flowchart of the proposed aneurysm detection method. 15
[Figure 3] Schematic illustration of the three sets of experiments. 17
[Figure 4] Visual results on meshes from MRA and CTA cases. 21
[Figure 5] Performance results for each test of the proposed model. 23