It is common for patients to undergo a sequence of imaging studies in the course of treatment ? e.g., in CT- based lung nodule surveillance, which has been found to reduce mortality in high-risk patients in the National Lung Screening Trial. Conventional imaging paradigms treat each scan in isolation, ignoring the tremendous amount of anatomical information shared between successive studies. Recently developed advanced reconstruction techniques challenge the conventional paradigm by incorporating prior image knowledge and yielding improved image quality and/or reduced radiation dose. Preliminary data on prior-image-based reconstruction (PIBR) methods applied to follow-up studies in pulmonary nodule surveillance suggest that radiation exposure may be reduced by an order-of-magnitude over current low-dose protocols. While encouraging, such PIBR algorithms require two main elements that would facilitate their clinical adoption: 1) an adaptation of PIBR to the lung nodule surveillance application including integration of accurate physical CT system models and sophisticated deformable registration to account for motion between sequential studies; and 2) a framework for understanding and controlling the propagation of prior information through the PIBR image formation process in a manner that permits consistent and reliable imaging of lung nodules. Current understanding of PIBR methods is limited ? while PIBR images qualitatively have high quality, traditional assessments have failed to address the fundamental differences in the biases associated with PIBR that differ from other reconstruction approaches. Thus, a quantitative analysis using clinically relevant metrics of image quality is required to ensure predictable PIBR behavior. We propose to achieve these goals through the following specific aims:
Aim 1 : Adapt PIBR to CT-based lung nodule surveillance: Develop an improved acquisition model for very low exposure CT including readout noise and develop sophisticated deformable prior image registration that accommodates non-anatomical change differences (e.g., varying inhalation states, positioning, etc.) between initial and follow- up scans.
Aim 2 : Construct a mathematical framework for analyzing prior image reconstructions to quantify and control the influence of prior data for accurate change visualization. Develop local approximations for PIBR analysis that enables 1) system analysis based on newly developed metrics of (spatially varying) admission-of-change; and 2) information source mapping that quantifies the extent to which image features are based on prior information or newly acquired data. These tools will be used to characterize and quantify the relationship between regularization and accurate reconstruction of change and to prospectively proscribe imaging parameters to reliably visualize specific levels of change. Successful completion of these aims offers to break conventional paradigms of image acquisition and reconstruction and maximize information usage in sequential tomographic studies, provide a new level of possible exposure reductions in sequential CT imaging, and create a quantitative framework for controlling the influence of prior data on 3D image quality.
It is common for patients to undergo a number of imaging studies throughout the course of diagnosis, treatment, and follow-up; including scenarios like CT-based lung nodule surveillance where suspicious lesions are re-imaged after a period of several month. While the traditional clinical paradigm treats sequential imaging studies in isolation, neglecting a wealth of patient-specific anatomical information, we propose to develop a new imaging workflow that leverages information from previous scans into the image formation processing for subsequent scans. This methodology permits dramatic reductions in the required radiation exposures for CT, and the proposed effort adapts the technique for lung nodule surveillance, characterizes the low dose limits, and developed a mathematical understand of how to control and optimize the new data processing technique.
| Zhang, Hao; Gang, Grace J; Dang, Hao et al. (2018) Prospective Image Quality Analysis and Control for Prior-Image-Based Reconstruction of Low-Dose CT. Proc SPIE Int Soc Opt Eng 10573: |
| Zhang, Hao; Gang, Grace J; Dang, Hao et al. (2018) Regularization Analysis and Design for Prior-Image-Based X-Ray CT Reconstruction. IEEE Trans Med Imaging 37:2675-2686 |
