Over the past 40 years, diagnostic medical imaging has undergone remarkable advancements in CT, MRI, and ultrasound technology. Today, the field is experiencing a major paradigm shift, thanks to significant and rapid progress in deep learning techniques. As a result, numerous innovative AI-based programs have been developed to improve image quality and enhance clinical workflows, leading to more efficient and accurate diagnoses.
AI advancements of medical imaging not only address existing unsolved problems but also present new and complex challenges. Solutions to these challenges can improve image quality and reveal new information currently obscured by noise, artifacts, or other signals. Holistic insight is the key to solving these challenges. Such insight may lead to a creative solution only when it is based on a thorough understanding of existing methods and unmet demands.
This book focuses on advanced topics in medical imaging modalities, including CT and ultrasound, with the aim of providing practical applications in the healthcare industry. It strikes a balance between mathematical theory, numerical practice, and clinical applications, offering comprehensive coverage from basic to advanced levels of mathematical theories, deep learning techniques, and algorithm implementation details. Moreover, it provides in-depth insights into the latest advancements in dental cone-beam CT, fetal ultrasound, and bioimpedance, making it an essential resource for professionals seeking to stay up-to-date with the latest developments in the field of medical imaging.
Author(s): Jin Keun Seo
Series: Mathematics in Industry, 40
Publisher: Springer
Year: 2023
Language: English
Pages: 348
City: Singapore
Preface I
Preface II
Acknowledgements
Contents
Contributors
Acronyms
1 Nonlinear Representation and Dimensionality Reduction
1.1 Introduction
1.2 Mathematical Notations and Definitions
1.3 Linear Dimensionality Reduction
1.3.1 Canonical Representation
1.3.2 Fourier Encoding
1.3.3 Wavelet Encoding
1.3.4 General Wavelet Basis
1.3.5 Principal Component Analysis (PCA)
1.3.6 Regularization and Compressed Sensing
1.4 Autoencoder and Manifold Learning
1.4.1 Linear and Semi-linear Autoencoder
1.4.2 Convolutional AE (CAE)
1.4.3 Variational Autoencoders (VAEs)
1.5 Application of Auto-Encoder: Automatic 3D Cephalometric Annotation System
1.5.1 The Overall Process of the Automatic 3D Cephalometric Annotation System
1.5.2 Generating the 2D Image Containing 3D Geometric Cues
1.5.3 Finding only Some of the Landmarks that are Relatively Easy to Find
1.5.4 Learning a Low-Dimensional Latent Representation of mathfrakR
1.5.5 Detecting the Total Landmark Vector (mathfrakR) from the Fractional Information (mathfrakR)
1.5.6 Remarks
References
2 Deep Learning Techniques for Medical Image Segmentation and Object Recognition
2.1 Introduction
2.2 Segmentation Problem
2.3 Conventional Segmentation Methods
2.3.1 Thresholding Methods
2.3.2 Region-Growing Method
2.3.3 Energy-Based Deformable Models
2.4 Deep Learning-Based Segmentation Methods
2.4.1 Convolutional Neural Networks (CNNs)
2.4.2 Fully Convolutional Networks
2.4.3 U-net and M-net
2.4.4 Confidence Map
2.4.5 YOLO
2.4.6 Dentistry Application: 3D Tooth Segmentation from 3D CBCT Image
2.4.7 Remarks on Deep Learning Methods
References
3 Deep Learning for Dental Cone-Beam Computed Tomography
3.1 Introduction
3.2 Basics of CT
3.2.1 History of CT
3.2.2 Parallel-Beam CT: Basic Principles
3.2.3 Fan-Beam CT: Reconstruction Algorithm
3.2.4 Cone-Beam CT
3.2.5 Dental CBCT
3.3 Dental CBCT Artifacts: Beam-Hardening
3.3.1 Lambert–Beer Law and Beam-Hardening Artifacts
3.3.2 Effect of Sinogram Discrepancies
3.3.3 Mathematical Analysis of Streaking Artifacts
3.3.4 Data Consistency Conditions for CBCT
3.4 Methods for MAR
3.4.1 Conventional MAR Methods
3.4.2 Phantom for Image Quality Evaluation
3.5 Deep Learning-Based Image Enhancement of Low-Dose Dental CBCT
3.5.1 Dental CBCT Data Acquisition Protocol
3.5.2 Towards Metal Artifact Reduction in Low-Dose Dental CBCT
3.6 Generation of Synthetic Data for MAR Using Machine Learning
3.6.1 Semi-synthetic Data Generation
3.6.2 Simulated Projection Data with Simulated Dental Crowns and Implants, and Orthodontic Braces
3.6.3 GAN-Based Synthetic-to-Realistic Image Refinement
3.7 Discussion and Conclusion
References
4 Artificial Intelligence for Digital Dentistry
4.1 Introduction
4.2 Development of AI-Based Data Integration Platform for Digital Dentistry
4.2.1 Traditional Versus Digital Dentistry
4.2.2 Necessity and Usefulness of AI-Based Digital Platform Integrating 3D Jaw–Teeth–Face Data
4.3 Individual Tooth Segmentation in IOS
4.3.1 Tooth Feature-Highlighted 2D Image Generation
4.3.2 Tooth Bounding Box Detection and 3D Tooth ROI Extraction Using Generated 2D Images
4.3.3 3D Segmentation for Individual Teeth from the 3D Tooth ROIs
4.4 A Fully Automated Method for 3D Individual Tooth Identification and Segmentation in Dental CBCT
4.4.1 Generation of Panoramic Images of the Upper and Lower Jaws from a 3D CBCT Image
4.4.2 Individual Tooth Detection, Identification and Segmentation in the 2D Reconstructed Panoramic Images
4.4.3 3D Segmentation for Individual Teeth from the 3D Tooth ROIs
4.5 Accurate Digital Impression Extraction Method of the Entire 3D Tooth Using CBCT and Intraoral Scanner
4.5.1 Rigid Transformation
4.5.2 Paired Point Methods
4.5.3 Removing Non-overlapping Parts Between X and Y that Can Affect ICP Registration
4.6 Discussion and Future Research Direction
References
5 Artificial Intelligence for Fetal Ultrasound
5.1 Introduction
5.2 Basics of Fetal Ultrasound
5.2.1 Sound Wave
5.2.2 Transabdominal and Transvaginal Ultrasound
5.2.3 Principle of Two-Dimensional (2D) B-mode Ultrasound Imaging
5.3 Ultrasound Artifacts
5.3.1 Interactions of Ultrasound with Tissues and Imaging Artifact
5.3.2 Reflection and Reverberation Artifact
5.3.3 Refraction and Edge Shadowing Artifact
5.3.4 Scattering and Speckle Artifact
5.3.5 Attenuation, Acoustic Shadowing Artifact and Acoustic Enhancement Artifacts
5.4 Fetal Ultrasound Measurements in List
5.4.1 Evaluation of Fetal Central Nervous System Malformations
5.4.2 Abdominal Circumference Measurement
5.4.3 Assessment of Amniotic Fluid Index
5.4.4 Measurement of Cervical Length
5.4.5 Other Fetal Examinations
5.4.6 Remarks
5.5 Deep Learning Methods for Fetal Ultrasound Measurement
5.5.1 DL-Based Automatic US Examination of the Fetal Central Nervous System
5.5.2 Discussion
5.6 Transvaginal Ultrasound: Automatic Measurement of Cervical Length
5.6.1 U-Net Based Models Without the Aid of Supplementary Learning of CL-related Features
5.6.2 Deep Learning Models with the Aid of Supplementary Learning of CL-Related Features
5.7 Discussion
References
6 Electrical Impedance Imaging
6.1 Introduction
6.2 Electrical Impedance Tomography (EIT)
6.2.1 Conductivity and Permittivity
6.2.2 Forward Problem in EIT
6.2.3 Inverse Problem in EIT
6.2.4 Sensitivity Analysis
6.2.5 Deep Learning-Based EIT
6.2.6 Applications
6.3 Magnetic Resonance Electrical Impedance Tomography (MREIT)
6.4 Electrical Properties Tomography
6.5 Discussion and Conclusion
References
7 Deep Learning for Ill Posed Inverse Problems in Medical Imaging
7.1 Introduction
7.2 Undersampled Magnetic Resonance Imaging (MRI)
7.2.1 MR Physics
7.2.2 Towards Highly Undersampled MRI
7.2.3 Deep Learning Approach
7.3 Discussion
References
