Medical Image Synthesis. Methods and Clinical Applications

Cover
Half Title
Series Page
Title Page
Copyright Page
Table of Contents
List of Contributors
Introduction
Section I: Methods and Principles
	Chapter 1: Non-Deep-Learning-Based Medical Image Synthesis Methods
		1.1 Introduction
		1.2 Overview of Non-learning-based Methods
			1.2.1 Single-Atlas-Based Method
			1.2.2 Multi-atlas-based Method
			1.2.3 Patch-based Atlas Method
		1.3 Overview of Traditional Machine-Learning-based Methods – Voxel-based Techniques
		1.4 Discussion
			1.4.1 Achievements
			1.4.2 Limitations
		1.5 Conclusion
		References
	Chapter 2: Deep-Learning-Based Medical Image Synthesis Methods
		2.1 Introduction
		2.2 Literature Searching
		2.3 Network Architecture
			2.3.1 NN
			2.3.2 CNN
			2.3.3 FCN
			2.3.4 GAN
				2.3.4.1 Conditional GAN
					2.3.4.1.1 DCGAN
					2.3.4.1.2 Pix2pix
					2.3.4.1.3 InfoGAN
				2.3.4.2 Cycle-GAN
					2.3.4.2.1 Res-Cycle-GAN
					2.3.4.2.2 Dense-Cycle-GAN
					2.3.4.2.3 Unsupervised Image-to-Image Translation Networks (UNIT)
					2.3.4.2.4 Bicycle-GAN
					2.3.4.2.5 StarGAN
			2.3.5 Loss Function
				2.3.5.1 Image Distance Loss
				2.3.5.2 Histogram Matching Loss
				2.3.5.3 Perceptual Loss
				2.3.5.4 Discriminator Loss
				2.3.5.5 Adversarial Loss
		2.4 Applications
			2.4.1 Multimodality MRI Synthesis
			2.4.2 MRI-only Radiation Therapy Treatment Planning
			2.4.3 CBCT Improvement/Enhancement
			2.4.4 Low-count PET and PET Attenuation Correction
		2.5 Summary and Discussion
		Disclosures
		References
Section II: Applications of Inter-Modality Image Synthesis
	Chapter 3: MRI-Based Image Synthesis
		3.1 Introduction
			3.1.1 Synthetic CT Quality
			3.1.2 MR-only Radiation Therapy
			3.1.3 PET Attenuation Correction
			3.1.4 Discussion
		References
	Chapter 4: CBCT/CT-Based Image Synthesis
		4.1 Synthetic CT from CBCT Images
		4.2 Synthetic MRI from CT/CBCT Images
		4.3 Synthetic DECT from Single-Energy CT
		4.4 Discussion
		References
	Chapter 5: CT-Based DVF/Ventilation/Perfusion Imaging
		5.1 Introduction
		5.2 CT-based DVF Imaging (CTDI)
			5.2.1 Conventional CTDI Methods
			5.2.2 Deep Learning-based CTDI Methods
				5.2.2.1 Supervised DVF Synthesis Network
				5.2.2.2 Unsupervised DVF Synthesis Network
		5.3 CT-based Ventilation Imaging (CTVI)
			5.3.1 Classical DIR-based CTVI Methods
				5.3.1.1 HU-based Method
				5.3.1.2 Volume-based Method
			5.3.2 Improved DIR-based CTVI Methods
				5.3.2.1 Hybrid Method
				5.3.2.2 Biomechanical Model
				5.3.2.3 Integrated Jacobian Formulation (IJF) Method
				5.3.2.4 Mass-Conserving Volume Change (MCVC) Method
				5.3.2.5 Multilayer Supervoxels Estimation Method
			5.3.3 Other CTVI Methods
				5.3.3.1 Attenuation Method
				5.3.3.2 Deep Learning Method
		5.4 CT-based Perfusion Imaging (CTPI)
			5.4.1 DIR-based CTPI Methods
			5.4.2 Deep Learning-based CTPI Methods
			5.4.3 CTPI Techniques for Other Anatomies
		5.5 Clinical Applications of CT-based DVF/Ventilation/Perfusion Imaging
		5.6 Concluding Remarks
		References
	Chapter 6: Imaged-Based Dose Planning Prediction
		6.1 Introduction
		6.2 Status
		6.3 Current Challenges and Future Perspectives
		References
Section III: Applications of Intra-Modality Image Synthesis
	Chapter 7: Medical Imaging Denoising
		7.1 Introduction
		7.2 Review of Medical Image Denoising Applications
			7.2.1 Image Denoising Problem Statement
			7.2.2 Denoising Methods
				7.2.2.1 Classical (Spatial Domain) Denoising Method
					7.2.2.1.1 Spatial Domain Filtering
					7.2.2.1.2 Variational Denoising Methods
						7.2.2.1.2.1 Total Variation (TV) Regularization
						7.2.2.1.2.2 Nonlocal Regularization
						7.2.2.1.2.3 Sparse Representation
						7.2.2.1.2.4 Low-rank Minimization
				7.2.2.2 Transform Techniques
					7.2.2.2.1 Transform Domain Filtering
					7.2.2.2.2 Data-adaptive Transform
						7.2.2.2.2.1 Independent Component Analysis (ICA)
						7.2.2.2.2.2 Principal Component Analysis (PCA)
					7.2.2.2.3 Non-data-adaptive Transform
						7.2.2.2.3.1 Spatial-frequency Domain
						7.2.2.2.3.2 Wavelet Domain
					7.2.2.2.4 Block-Matching and 3D (BM3D) Filtering
				7.2.2.3 Deep Learning (DL)-based Image Denoising Methods
		7.3 Convolutional Neuron Network (CNN) for Medical Image Denoising and Resolution Restoration
		7.4 Different Medical Image Modalities
			7.4.1 Computed Tomography (CT)
				7.4.1.1 Reconstruction
					7.4.1.1.1 Filtered Back Projection
					7.4.1.1.2 Iterative Reconstruction
			7.4.2 Magnetic Resonance Imaging (MRI)
				7.4.2.1 MRI Denoising Approaches
			7.4.3 Positron Emission Tomography
				7.4.3.1 PET Denoising Approaches
			7.4.4 Ultrasound
				7.4.4.1 US Denoising Approaches
		7.5 Deep Learning Approaches for CT Denoising
			7.5.1 Convolutional Neuron Network Approached Design Optimization for Medical Image Denoising
			7.5.2 STIR-Net: Spatial-Temporal Image Restoration Net for CT Perfusion Radiation Reduction
		7.6 Summary and Discussion
		Acknowledgment
		References
	Chapter 8: Attenuation Correction for Quantitative PET/MR Imaging
		8.1 Introduction
		8.2 Methods for PET Attenuation Correction
			8.2.1 Atlas-based Methods
			8.2.2 Segmentation-based Methods
			8.2.3 Joint Estimation-based Methods
			8.2.4 Machine Learning-based Methods
			8.2.5 Deep Learning-based Methods
		8.3 Conclusion
		References
	Chapter 9: High-Resolution Image Estimation using Deep Learning
		9.1 Introduction
		9.2 Methods
			9.2.1 Self-supervised Learning Framework
			9.2.2 The cycleGAN
		9.3 Results
			9.3.1 Ultrasound High-resolution Image Estimation
				9.3.1.1 Breast US
				9.3.1.2 Prostate US
			9.3.2 CT High-resolution Image Estimation
			9.3.3 MRI High-resolution Image Estimation
		9.4 Discussion
		9.5 Conclusion
		References
	Chapter 10: 2D–3D Transformation for 3D Volumetric Imaging
		10.1 Introduction
		10.2 Methods
			10.2.1 Overview of the Deep-Learning-Based 2D–3D Transformation Methods
			10.2.2 Network Training
			10.2.3 Supervision and Loss Function
		10.3 Evaluation Results
		10.4 Discussion
		References
	Chapter 11: Multimodality MRI Synthesis
		11.1 Introduction
		11.2 Multimodality MRI Synthesis via Patch-Based Conventional Learning
			11.2.1 A Dual-Domain Cascaded Regression for 7T MRI Prediction from 3T MRI
			11.2.2 An Evaluation of Different Patch-Based Conventional Learning Methods
		11.3 Multimodality MRI Synthesis via Deep Learning
			11.3.1 A Fully Supervised 7T MRI Prediction from 3T MRI with CNNs
				11.3.1.1 Method
				11.3.1.2 Performance Evaluation
			11.3.2 A Semi-Supervised Adversarial Learning for 7T MRI Prediction from 3T MRI
				11.3.2.1 Method
				11.3.2.2 Performance Evaluation
		11.4 Conclusion
		References
	Chapter 12: Multi-Energy CT Transformation and Virtual Monoenergetic Imaging
		12.1 Introduction
		12.2 Implementation of Multi-energy CT Imaging
			12.2.1 Sequential Scanning
			12.2.2 Fast Tube Potential Switching
			12.2.3 Multilayer Detector
			12.2.4 Dual-Source Acquisitions
			12.2.5 Beam Filtration Techniques
			12.2.6 Energy-Resolved Detector
		12.3 Motivation of Artificial Intelligence-based Multi-energy CT Imaging
		12.4 Energy Domain MECT Image Synthesis
		12.5 Cross-Domain MECT Image Synthesis
		12.6 Remaining Challenges and Future Work
		12.7 Summary
		Acknowledgment
		References
	Chapter 13: Metal Artifact Reduction
		13.1 Introduction
		13.2 Review of MAR in Different Modalities
			13.2.1 MAR in Magnetic Resonance Imaging (MRI)
			13.2.2 MAR in Computed Tomography (CT)
		13.3 Supervised Dual-domain Learning for MAR
			13.3.1 Overview of Dual-domain Learning Framework
			13.3.2 Generation of the Prior Image
			13.3.3 Deep Sinogram Completion
			13.3.4 Objective Function
			13.3.5 Dataset and Implementation
			13.3.6 Experimental Results on DeepLesion Data
				13.3.6.1 Quantitative Comparisons
				13.3.6.2 Qualitative Analysis
			13.3.7 Generalization to Different Site Data
			13.3.8 Experiments on CT Images with Real Metal Artifacts
				13.3.8.1 Results on Real Metal-Corrupted CT Images
				13.3.8.2 The Influence of Metal Segmentation
			13.3.9 Analysis of Our Approach
				13.3.9.1 Effectiveness of Prior Image Generation
				13.3.9.2 Effectiveness of Residual-Sinogram-Learning
				13.3.9.3 Compared with Tissue Processing
		13.4 Self-Supervised Dual-Domain MAR
			13.4.1 Overview
			13.4.2 Self-Supervised Cross-Domain Learning
				13.4.2.1 Sinogram Completion
				13.4.2.2 FBP Reconstruction Loss
				13.4.2.3 Image Refinement
			13.4.3 Prior-Image-Based Metal Trace Replacement
			13.4.4 Training and Testing Strategies
			13.4.5 Datasets and Implementation
			13.4.6 Ablation Study
			13.4.7 Comparison with Other Methods
			13.4.8 Qualitative Analysis
			13.4.9 Experiments on CT Images with Real Artifacts
			13.4.10 Experiments on Different Site Data
			13.4.11 Comparison with More Recent Methods
		13.5 Discussion and Conclusion
		References
Section IV: Other Applications of Medical Image Synthesis
	Chapter 14: Synthetic Image-Aided Segmentation
		14.1 Introduction
		14.2 Modality Enhancement-Based Segmentation
			14.2.1 Multi-Modality Image Synthesis
			14.2.2 Similar-Modality Image Synthesis
		14.3 Training Data Enlargement-Based Segmentation
		14.4 Summary and Discussion
		Disclosures
		References
	Chapter 15: Synthetic Image-Aided Registration
		References
	Chapter 16: CT Image Standardization Using Deep Image Synthesis Models
		16.1 Introduction
		16.2 Background
			16.2.1 CT Image Acquisition and Reconstruction Parameters
			16.2.2 Radiomic Features
			16.2.3 Deep Generative Models for Image Synthesis
				16.2.3.1 U-net
				16.2.3.2 Generative Adversarial Network
		16.3 CT Image Standardization Model
			16.3.1 CNN-based CT Image Standardization
			16.3.2 GAN-based CT Image Standardization
				16.3.2.1 GANai
				16.3.2.2 STAN-CT
				16.3.2.3 RadiomicGAN
				16.3.2.4 CVH-CT
		16.4 Discussion and Conclusion
		References
Section V: Clinic Usage of Medical Image Synthesis
	Chapter 17: Image-Guided Adaptive Radiotherapy
		17.1 Introduction
		17.2 CBCT-based Online Adaptive Radiation Therapy System
		17.3 MR-Guided Real-Time Adaptive Radiation Therapy
		17.4 CBCT-Guided Adaptive Proton Therapy
		17.5 Online Adaptive RT Using Plan-of-the-Day
		17.6 Dose Gradient Adaptation
		References
Section VI: Perspectives
	Chapter 18: Validation and Evaluation Metrics
		18.1 Introduction
		18.2 Overview of Qualitative Validation
		18.3 Overview of Quantitative Validation
			18.3.1 Similarity Measures
			18.3.2 Dice Measures
			18.3.3 Dosimetric Agreement
		References
	Chapter 19: Limitations and Future Trends
		References
Index