The goal of this research was to propose and evaluate solutions to four important aspects of adaptive radiation therapy in order to make it more reliable, accurate, and efficient in clinical environment. The first study focused on the evaluation of several deformable image registration algorithms. Results demonstrated that the Dense Anatomical Block Matching registration outperformed the other methods making it a very promising alternative to the existing registration methods for challenging CT-to-CBCT registration and its applications for radiation dose calculation, dose mapping and contour propagation in adaptive radiation therapy (ART) of the pelvic region. The second study focused on the quantitative evaluation of eight proposed adaptive radiation therapy approaches for prostate cancer patients treated with hypofractionated VMAT. The ART strategies included online and offline methods. The comprehensive analysis showed that daily on-line adaptation approaches were the most impactful. The findings of this study provided applicable insights into the selection of the optimal ART strategy, improving the quality of the decision-making process based on the quantitatively evaluated dosimetric benefits. The third study aimed to utilize a deep learning network to automatically contour critical organs on the computed tomography (CT) scans of head and neck cancer patients. Proposed model achieved expert level accuracy and was able to segment 25 critical organs on unseen CT images in approximately 7 seconds per patient. High accuracy and short contouring time allow for the implementation of the model within a clinical ART workflow, which would lead to a significant decrease in the time required to create a new adapted treatment plan. The objective of the fourth study was to use artificial intelligence methods to build a decision making support system that would classify previously delivered plans of brain tumor patients into those that met treatment planning objectives and those for which objectives were not met due to the priority given to one or more organs-at-risk. Among evaluated machine learning algorithms, the Logistic Regression model achieved the highest accuracy and can be used by radiation oncologists to support their decision-making process in terms of treatment plan adaptations and plan approvals in a data-driven quality assurance program.

This book, now in an extensively revised and updated second edition, provides a comprehensive overview of both machine learning and deep learning and their role in oncology, medical physics, and radiology. Readers will find thorough coverage of basic theory, methods, and demonstrative applications in these fields. An introductory section explains machine and deep learning, reviews learning methods, discusses performance evaluation, and examines software tools and data protection. Detailed individual sections are then devoted to the use of machine and deep learning for medical image analysis, treatment planning and delivery, and outcomes modeling and decision support. Resources for varying applications are provided in each chapter, and software code is embedded as appropriate for illustrative purposes. The book will be invaluable for students and residents in medical physics, radiology, and oncology and will also appeal to more experienced practitioners and researchers and members of applied machine learning communities.

ABSTRACT: Image guided radiation therapy (IGRT) requires developing advanced methods for target localization. Once target motion is identified, the patient specific treatment margin can be incorporated into the treatment planning, accurately delivering the radiation dose to the target and minimizing the dose to the normal tissues. Deformable image registration (DIR) has become an indispensable tool to analyze target motion and measure physiological change by temporal imaging or time series volumetric imaging, such as four-dimensional computed tomography (4DCT). Current DIR algorithms suffer from inverse inconsistency, where the deformation mapping is not unique after switching the order of the images. Moreover, long computation time of current DIR implementation limits its clinical application to offline analysis. This dissertation makes several major contributions: First, an inverse consistent constraint (ICC) is proposed to constrain the uniqueness of the correspondence between image pairs. The proposed ICC has the advantage of 1) improving registration accuracy and robustness, 2) not requiring explicitly computing the inverse of the deformation field, and 3) reducing the inverse consistency error (ICE). Moreover, a variational registration model, based on the maximum likelihood estimation, is proposed to accelerate the algorithm convergence and allow for inexact image pixel matching within an optimized variation for noisy image pairs.

​This book provides a complete overview of the role of machine learning in radiation oncology and medical physics, covering basic theory, methods, and a variety of applications in medical physics and radiotherapy. An introductory section explains machine learning, reviews supervised and unsupervised learning methods, discusses performance evaluation, and summarizes potential applications in radiation oncology. Detailed individual sections are then devoted to the use of machine learning in quality assurance; computer-aided detection, including treatment planning and contouring; image-guided radiotherapy; respiratory motion management; and treatment response modeling and outcome prediction. The book will be invaluable for students and residents in medical physics and radiation oncology and will also appeal to more experienced practitioners and researchers and members of applied machine learning communities.

Modern medical imaging and radiation therapy technologies are so complex and computer driven that it is difficult for physicians and technologists to know exactly what is happening at the point-of-care. Medical physicists responsible for filling this gap in knowledge must stay abreast of the latest advances at the intersection of medical imaging an