DDespite its long history and established clinical utility, X-ray Computed Tomography (CT) remains an area of active research and innovation. In the recent years, significant progress has been made in scanner hardware, reconstruction algorithms, and dose reduction techniques. The technical development is paralleled by continuous expansion of diagnostic applications of CT. In this session, we will present an overview of some of these important advances, including (1) high-resolution CT systems, (2) personalized CT dosimetry, (3) new clinical capabilities in assessment of lung disease and (4) learning-based reconstruction algorithms.
One of the areas of major current progress in CT hardware has been in improving the spatial resolution of whole-body CT. The first talk will present comprehensive measurements of technical and clinical performance of one of the recently introduced high-resolution CT scanners. The system has six user-selectable focal spots, and three different detector configurations for acquisition. The MTF measurements in both the in-plane (x, y) and longitudinal (z) axes demonstrate a true doubling in spatial resolution. Clinical applications which benefit from the resolution of this scanner include thoracic, small vessel, pediatric, and musculoskeletal imaging.
Advances in computational and deep learning technologies are enabling rapid, patient-specific CT organ dose estimation. Patient-specific organ dose estimation uses the CT images from the patient's scan to develop a personalized phantom model and to accurately model patient-specific aspects of the CT scan that affect dose, such as tube current modulation and patient positioning. Patient-specific organ dose estimation is typically performed through two steps: (1) estimating a dose distribution map and (2) determining organ boundaries that are then applied to the dose map. Monte Carlo, Linear Boltzmann Transport Equation (LBTE) solver, and deep learning methods have been developed for rapid dose map estimation. Deep learning methods have also been applied in organ segmentation. The second talk will present an overview of patient-specific CT organ dose estimation methods and potential applications. The development and validation of a specific tool, using a rapid LBTE solver and deep learning segmentation methods, will be described.
The third talk will focus on recent advances in clinical application of CT. With careful attention to image protocol details, CT and Dual Energy CT now provides for a detailed map of parenchyma, vessels and airways along with a characterization of pathophysiology. Image matching between multiple lung volumes yields regional metrics related to lung mechanics and provides a marker of underlying functional small airways disease. Dual Energy-based material decomposition of contrast enhanced images provide a measure of regional perfused blood volume which serves as a surrogate for regional perfusion. With regional Jacobians from image matching providing regional measures of ventilation and DECT-PBV providing regional measures of perfusion, one can now assess the matching of ventilation and perfusion which is the primary function of the lung, required for gas exchange. The application of these CT metrics to the understanding of the etiology of smoking-associated lung disease will be discussed.
The session will conclude with a presentation of the emerging learning-based methods that exploit datasets of regular-dose and/or low-dose CT scans for improved image reconstruction. The talk will first review an approach based on model-based image reconstruction (MBIR), where the sparsity-based regularizer parameters are learned in an unsupervised manner from a dataset of high-quality CT images. A union of sparsifying transforms is pre-learned to cluster CT image patches into multiple groups, with a specific transform well-matched to each group. When incorporated as a data-driven regularizer in MBIR, the learned transforms provide much better image reconstruction than conventional filtered backprojection and non-adaptive regularization methods especially at low X-ray doses. Next, an extension of this approach to a unified supervised-unsupervised (SUPER) learning scheme will be discussed. This scheme combines classical MBIR optimization and unsupervised transform learning together with recent supervised deep learning in a common framework. Multiple interpretations of this unified framework will be discussed and shown to provide much better image reconstructions than the constituent models with limited training data.
1) Understand the performance characteristics of high-resolution whole-body CT
2) Understand the computer modeling techniques that enable personalized CT dose estimation
3) Understand the advanced CT techniques used in assessment of lung function
4) Understand the unsupervised and supervised-unsupervised approaches for learning-based CT image reconstruction.
Funding Support, Disclosures, and Conflict of Interest: WZ receives research support from Siemens Healthineers