Room: Exhibit Hall | Forum 9
Purpose: Katsevich algorithm (KAT) is a theoretically exact analytic reconstruction method for helical CT. This work develops a novel method that iteratively utilizes KAT, namely iterative Katsevich algorithm (IKAT).
Methods: IKAT is a modified proximal forward-backward splitting algorithm. Each iteration consists of three steps: first the current image is projected to the data domain to form a data residual; second this data residual is inversed by KAT to a difference image and then weighted with the current image to form an intermediate image, last this intermediate image is denoised to form the final image for the current iteration. projection operator, IKAT converges in much fewer iterations than standard iterative reconstruction (IR) method.
Results: The reconstructed image quality was evaluated with Catphan scans on a Siemens Biograph mCT scanner. The spatial resolution was quantitatively computed as the average of MTF for 3 to 5 lp/mm, and qualitatively compared for 6 to 7 lp/mm; the image contrast was quantified as the average of CNR for medium-contrast objects (PMP: -200HU; Delin: 340HU; LDPE: 100HU) and low-contrast objects (Polystryrene: -35HU; periphery 1% contrast: 10HU). Since iterative methods are subject to choices of reconstruction parameters, the multi-parameter MTF v.s. CNR tradeoff curves were plotted for rigorous comparison between IR and IKAT. With similar CNR values, (CNR, MTF) was (20, 0.14), (14, 0.10), and (27, 0.26) for the reference, KAT, IR, and IKAT respectively.
Conclusion: IKAT improves both image quality and reconstruction speed from IR and provides better image quality than the reconstructed images from the CT scanner that are used clinicall.