(Thursday, 8/2/2018) 1:00 PM - 3:00 PM
Room: Room 202
Purpose: Prior-image-based reconstruction (PIBR) has demonstrated great potential to reduce CT radiation dose beyond traditional limits. One challenge with PIBR is how to control the amount of information integrated from the prior image. Too little prior information yields marginal benefits, while too much information can obscure important anatomical changes (e.g., differences in the appearance of a pulmonary nodule between sequential studies). In this work we present a theoretical framework for predicting and quantifying reconstruction bias. This framework permits prospective regularization control for reliable and robust reconstruction performance.
Methods: We demonstrate that prior-image-based regularization introduces a reconstruction bias related to the change between the prior image and the current anatomy. We quantify this bias and derive an analytical relationship between regularization strength and the fractional contrast change between the prior and the reconstruction. This relationship serves as a tool for predicting PIBR bias properties and prospectively determining regularization strength for user-specified levels of contrast differences. We validate the efficacy of the proposed framework using cadaver and National Lung Screening Trial datasets in scenarios of nodule appearance and growth cases.
Results: In comparisons between the proposed analytical prediction approach and brute force exhaustive evaluations based on iterative reconstruction across regularization strengths, we find the proposed method is highly accurate in predicting biases in the reconstruction of lung nodule features in sequential studies. Moreover, we demonstrate that these bias predictions may be used to enforce reliable reconstruction of lung nodule changes based on user-defined bias criteria.
Conclusion: We have proposed and validated a novel prediction framework enabling quantification of the relationship between image features and PIBR regularization strength. This framework enables robust and reliable PIBR, and is a critical breakthrough that facilitates clinical adoption of PIBR in lung nodule surveillance and other sequential imaging studies.
Funding Support, Disclosures, and Conflict of Interest: This research was supported in part by an academic-industry partnership with Elekta AB (Stockholm, Sweden), NIH grant R21 CA219608, and an AAPM Research Seed Funding Grant.
Keywords
Not Applicable / None Entered.
Taxonomy
Contact Email
