Core C - Computational and Quantitative Imaging Core

University Of Tx Md Anderson Can Ctr, Houston TX

Investigators

Abstract

ABSTRACT: Computational and Quantitative Imaging (Core C) Advances in imaging have improved our understanding of tumor heterogeneity, normal tissue variability, and early detection of tumor response and normal tissue toxicity. The rich information content of these advanced imaging methods, combined with the growing capacity to collect multiple types of images from various sources and of various dimensions (2D, 3D, 4D) and time points during therapy opens an exciting prospect for image- based guidance and assessment of clinical interventions. A major obstacle to full exploitation of this paradigm is the natural deformations (i.e., positional differences in soft tissue, deformation during swallowing and breathing, and morphologic soft tissue change) in anatomy between imaging events or between imaging and intervention events. This obstacle must be removed to harness and leverage the information presented in the data. The Computational and Quantitative Imaging Core will further optimize and deploy anatomical models incorporating 2D information into 3D models will quantify the complex physiologic changes of the patient over time. Further, we will enable novel deep learning methods to harness this information into toxicity prediction models which will allow us to reduce the severity and frequency of toxicity events for future patients. The innovative, multi- disciplinary development and deployment proposed here will advance these anatomical models to facilitate high precision, localized, image guided cancer therapy in the head and neck. These tools will be focused on the following specific aims: Aim 1: Automated Segmentation and Image Processing Algorithms. Leveraging our ongoing, successful work in tumor and normal tissue segmentation on CT and MR, we will expand to include comprehensive segmentation tools for all organs, tumors, and structures of interest in the head and neck on CT, MR, fluoroscopy, and optical imaging in support of correlative pathology. Aim 2: 2D/3D/4D Biomechanical Model-based Deformable Image Registration. This suite of tools will include 2D (fluoroscopy, cine, and histopathology imaging), 3D (CT, MR, and optical images obtained during correlative pathology), and 4D (multi-time point and dynamic imaging) image registration across all imaging modalities as well as the incorporation of dose throughout the registration process. Aim 3: Deep Learning Prediction Models. Interpretable deep learning and traditional machine learning algorithms are critical to support the toxicity and outcomes prediction models planned in all projects.