Program Information
BEST IN PHYSICS (IMAGING): Task-Based Parameter Optimization for Low Signal Correction in Low Dose CT
D Gomez-Cardona*, J Hayes , R Zhang , K Li , G Chen , University of Wisconsin-Madison, Madison, WI
Presentations
TH-AB-601-11 (Thursday, August 3, 2017) 7:30 AM - 9:30 AM Room: 601
Purpose: Low signal correction (LSC) methods can reduce noise streaks and noise level in CT to provide high quality images at low dose levels. Parameter selection for LSC requires many trial-and-error experiments often leading to variable performance for different imaging tasks. The purpose of this work is to develop a task-based parameter optimization framework applicable to a given LSC method.
Methods: Since LSC methods reduce noise in the raw data locally, noise and spatial resolution may not be uniform across the field of view (FOV). In the proposed framework, noise and spatial resolution are measured and incorporated into a model observer to study the performance across the FOV for a given imaging task. The detectability index is then used as figure of merit to optimize parameter selection in LSC. The framework was implemented for two LSC methods: adaptive trimmed mean (ATM) filter and anisotropic diffusion (AD) filter. Repeated CT scans of an anthropomorphic phantom were acquired at a reference dose and reduced dose. Spatial resolution (MTF₁₀) and noise texture (2DNPS isotropy) contours were obtained to illustrate the performance of the methods at each point in the parameter space. Finally, detectability maps for a given task were obtained to guide the selection of optimal parameters for both methods.
Results: The spatial resolution and noise texture contours depicted different trends across the parameter space for both methods. Detectability contours of a given task accommodated these trade-offs to match visual perception of the images. Near-optimal parameters were found for each method by meeting or exceeding prescribed detection performance criteria under the additional constraint that the visual perception of the image for other tasks was still good.
Conclusion: A task-based framework to optimize the selection of parameters for LSC was developed. This framework is generalizable to other LSC methods that can be parameterized.
Contact Email: