2187 - Improving CT Image Quality Outside of the Regular Scan Field of View for Radiation Therapy Planning Using a Novel Reconstruction Algorithm
Presenter(s)
G. Noid1, M. W. Straza Jr2, E. S. Paulson1, J. Shah3, M. Baer-Beck4, and A. Tai1; 1Department of Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, 2Department of Radiation Oncology, Froedtert & the Medical College of Wisconsin, Milwaukee, WI, 3Siemens Healthineers, Cary, NC, 4Siemens Healthineers, Forchheim, Germany
Purpose/Objective(s): Accurate patient anatomy is essential for radiation treatment planning. In computer tomography (CT), image quality can be dramatically degraded outside the design field of view (FOV), often resulting in i) reduced CT number and ii) increased likelihood to misrepresent the external patient surface. Uncertainty in either of these factors can introduce errors in radiotherapy dose calculations. A novel extended field of view reconstruction algorithm which is supported by a convolutional neural network (HDFOV 5.0) was developed to enable visualization of the human body parts and skin line located outside the scan field of view up to the bore size. In this work we investigated the ability of HDFOV 5.0 to improve image quality of extended FOV CT images compared to the current reconstruction algorithm HDFOV 4.0.
Materials/Methods: Images of 44 patients were acquired on Somatom Go.Open and Naeotom Alpha (Siemens Healthineers) CT scanners. The datasets were reconstructed using 2 different algorithms: extended FOV with HDFOV 4.0, and extended FOV with HDFOV 5.0. Three expert readers (2 medical physicists, one radiation oncologist) were presented with image data without further information in a randomized order across all datasets and blinded. The 130 data sets were presented both side-by-side and case-by-case. The readers were instructed to rate the images with 2 ratings: A Likert-scale for the anatomical representation outside the scan (design) FOV and a binary judgement whether or not the image can be considered for dose calculations.
Results: Overall, the readers preferred the HDFOV 5.0 data sets over the HDFOV 4.0 data sets. For the case-by-case comparison across all three readers the mean Likert-scale score was 3.07 with HDFOV 4.0 and 3.72 for HDFOV 5.0. For the side-by-side comparison across all three readers the mean Likert-scale score was 3.12 with HDFOV 4.0 and 3.59 for HDFOV 5.0. For binary judgement rating, mean dose planning ratings are 0.42 with HDFOV 4.0 and 0.65 with HDFOV 5.0 and 0.39 with HDFOV 4.0 and 0.67 with HDFOV 5.0 for case by case comparison and side by side comparison, respectively. These differences were statistically significant (with p<0.05) for both Likert-scale rating and binary judgement rating
Conclusion: Our reader study demonstrates that the new extended FOV reconstruction in HDFOV 5.0 improved CT image quality in comparison to the existing HDFOV 4.0 algorithm. While further investigation is merited, the HDFOV 5.0 algorithm is very promising for providing enhanced CT images for radiation therapy planning at the limits of reconstruction volumes.