Title Page
Abstract
Contents
1. Introduction 10
1.1. Radiomics 11
1.2. Reproducibility of radiomics 15
1.3. Multiparametric MRI based radiomics 22
1.4. Objectives 24
2. Study Ⅰ: Extraction of High Reproducible Radiomics Features Using Phantom 26
2.1. Methods 26
2.1.1. Phantom experiments 26
2.1.2. MRI acquisition and image preprocessing 29
2.1.3. ROI defines 29
2.1.4. Radiomics feature extraction and computational reproducibility 32
2.1.5. Construction of reproducible MRI radiomics features 33
2.2. Results 34
2.2.1. Intra-scanner repeatability (test-retest) of the radiomics features 34
2.2.2. Inter-scanner reproducibility of radiomics features 37
2.2.3. Computational reproducibility of radiomics features 37
3. Study Ⅱ: Clinical Retrial of High Reproducible radiomics features 42
3.1. Methods 42
3.1.1. Participants 42
3.1.2. MRI acquisition and image preprocessing 44
3.1.3. ROI segmentation 45
3.1.4. Features extraction 48
3.1.5. Classification model and statistical analysis 48
3.2. Results 50
3.2.1. Optimization of the radiomics models in the clinical data 50
3.2.2. Diagnostic performance of the radiomics model on the validation Set 50
4. Discussions 55
References 60
국문 요약 64
Table 1. Overview of studies on radiomics reproducibility using MRI. 19
Table 2. Information about Phantom construction of layers. 28
Table 3. Imaging acquisition parameters (phantom data). 31
Table 4. Test-Retest repeatability results of the phantom study: Pyradiomics. 35
Table 5. Test-Retest repeatability results of the phantom study: MITK. 36
Table 6. Multiple scanner repeatability and reproducibility results of the phantom study. 38
Table 7. Features showing high imaging within scanner repeatability and inter-scanner reproducibility, and high computational reproducibility from the phantom study. 40
Table 8. Patient Characteristics of the Study Population. 43
Table 9. Imaging acquisition parameters (clinical data). 45
Table 10. Performance of the reproducible and non-reproducible radiomics feature models in the patient groups. 52
Table 11. Statistical significance using DeLong's test for receiver operating characteristics analysis. 54
Figure 1. Flowchart of the radiomics analysis process. 13
Figure 2. Reproducibility in radiomics research. 16
Figure 3. Phantom and MRI data in our study. 28
Figure 4. Summary of the study design. 30
Figure 5. Reproducible radiomics features with high test-retest and inter-scanner reproducibility. 39
Figure 6. Reproducible radiomics features with both imaging and computational reproducibility. 41
Figure 7. The segmentation model for brain tumor using 3d-nnunet. 46
Figure 8. Example of patients' enhancing tumor ROI. 47
Figure 9. Schematic diagram of the overall modeling method. 49
Figure 10. Comparison of the diagnostic performances of the radiomics models using non-reproducible and reproducible radiomics features. 53