Title Page

Contents

Abstract 12

General Introduction 14

Chapter 1. Development and clinical validation of CT-based regional modified Centiloid method for amyloid PET 17

1.1. Introduction 17

1.2. Materials and methods 20

1.2.1. Participants 20

1.2.2. MRI data acquisition 21

1.2.3. Aβ PET-CT data acquisition 21

1.2.4. Regional visual assessment and rdcCL scales 23

1.2.5. Development of MRI-based global and regional CTX VOIs 23

1.2.6. Development of CT-based global and regional CTX VOIs 25

1.2.7. Development of MRI and CT-based rdcCL 26

1.2.8. Validation of the clinical efficacy of CT-based rdcCLs in the independent cohort 30

1.2.9. Statistical analysis 31

1.3. Results 33

1.3.1. Demographics of the participants 33

1.3.2. Visual assessment and rdcCL scales 36

1.3.3. CT-based rdcCLs 38

1.3.4. Reliability and precision in the MRI-based and CT-based rdcCLs 39

1.3.5. Validation of CT-based rdcCLs in independent participants 43

1.4. Discussion 46

1.5. Conclusions 50

1.6. Supplementary information 51

Chapter 2. Clinical and pathological validation of CT-based regional harmonization methods of amyloid PET 73

2.1. Introduction 73

2.2. Materials and methods 75

2.2.1. Participants 75

2.2.2. Aβ PET-CT imaging 78

2.2.3. Regional CT-based dcCL scales 78

2.2.4. Neuropathological assessment measurements 79

2.2.5. Validation of the clinical utility of CT-based dcCLs in an independent cohort 81

2.2.6. Statistical analysis 82

2.3. Results 84

2.3.1. Demographics of the participants 84

2.3.2. Relationships between regional dcCLs and Aβ plaques in each region 87

2.3.3. Optimal threshold values of regional dcCLs through Aβ pathological scores 90

2.3.4. Clinical utility of CT-based regional dcCLs in the independent cohort 92

2.4. Discussion 98

2.5. Conclusions 103

2.6. Supplementary information 104

Chapter 3. Deep learning-based quantitative analysis of Tau PET-CT imaging for Alzheimer's disease 106

3.1. Introduction 106

3.2. Materials and methods 109

3.2.1. Participants 109

3.2.2. MRI data acquisition 111

3.2.3. PET-CT data acquisition 111

3.2.4. Data preparation and image processing 112

3.2.5. Model architectures 113

3.2.6. Cascaded networks for Tau PET quantification using deep learning predictive models 116

3.2.7. Deep learning model evaluation for performance 120

3.2.8. Statistical analysis 120

3.3. Results 121

3.3.1. Demographics of the participants 121

3.3.2. Evaluation of segmentation performance 122

3.3.3. Evaluation of Tau PET quantification 126

3.4. Discussion 131

3.5. Conclusions 135

References 137

논문요약 152

Chapter 1 9

Table 1. Participant demographics and clinical findings in head-to-head and clinical validation cohorts 34

Table 2. Participant demographics and clinical findings of the subgroups 35

Table 3. Visual assessment and rdcCL scales 37

Chapter 2 9

Table 1. Clinical and pathological characteristics of participants 86

Table 2. Demographics and clinical features of subgroups in the clinical cohort 93

Chapter 3 9

Table 1. Performance measurements of brain mask segmentation 123

Table 2. Performance measurements of related Braak staging regions segmentation 125

Chapter 1 10

Fig. 1. Overview of the processing pipeline for regional Centiloid in CT-based and MRI-based methods 28

Fig. 2. Plots of correlation of global and regional rdcSUVR between MRI-based and CT-based methods for FMM 40

Fig. 3. Plots of correlation of FMM rdcCL between MRI-based and CT-based methods 41

Fig. 4. Bland-Altman plots of rdcCLs between MRI-based and CT-based Centiloid methods for FMM 42

Fig. 5. Comparison of the neuropsychological performance classified into four groups based on regional, global, and striatal cutoffs 44

Chapter 2 10

Fig. 1. Flow charts of clinical and pathological cohorts 77

Fig. 2. Plots of correlation between CT-based dcCL and regional CERAD scores (A-F) and relationships between striatum dcCL and striatal involvements of... 88

Fig. 3. ROC curves of regional dcCL scales and regional positivity classified by cutoff 91

Fig. 4. Comparison of the neuropsychological performances and hippocampal volume classified into four groups based on global and regional thresholds 95

Chapter 3 11

Fig. 1. Overview of the segmentation and quantification pipeline for the proposed deep cascaded networks 119

Fig. 2. Plots of correlation of SUVRs between SoT and three networks 128

Fig. 3. Bland-Altman plots of SUVRs between SoT and three networks 129