Program Information
Evaluation of CAD Nodule Detection Performance in Low Dose CT Lung Cancer Screening Across a Range of Dose Levels, Slice Thicknesses and Reconstruction Kernels
N Emaminejad*, M Wahi-Anwar, J Hoffman, A Sultan, K Ruchalski, G Kim, J Goldin, M Brown, M McNitt-Gray , UCLA School of Medicine, Los Angeles, CA
Presentations
WE-G-201-4 (Wednesday, August 2, 2017) 4:30 PM - 6:00 PM Room: 201
Purpose: Performing the lowest possible dose CT scan while still accomplishing the desired clinical task is an important goal, even in the context of CT lung cancer screening where doses are already low. The purpose of this work was to investigate the effects of dose reduction, and different slice thicknesses and reconstruction kernels on nodule detection performance by CAD.
Methods: Image data and raw projection data from 50 patients undergoing low dose CT for lung cancer screening were acquired. All scans were performed using a 64 slice MDCT with Tube Current Modulation (TCM) according to the AAPM lung cancer screening protocols. Radiologists reviewed the original image dataset and identified solid nodules approximately 4mm or larger. Simulated (further) reduced dose scans at 50%, 25%, and 10% of the original dose were created by adding calibrated noise to the projection data. The raw data was then reconstructed using slice thicknesses of 1 and 2 mm and two kernels (smooth and medium), resulting in 16 conditions for each patient dataset. Each image dataset was then input to an in-house CAD software for nodule detection. Nodule level sensitivity of CAD was measured and assessed across all conditions.
Results: Radiologists marked 8 solid nodules in our cohort; 2 cases with 2 nodules and 4 cases with 1 nodule. The size distribution was 4 nodules > 8mm in diameter, 3 nodules in 6 to 8mm range and 1 nodule < 6m. CAD nodule detection sensitivity was as high as 62.5% and was maintained across a wide range of dose levels and reconstruction kernels. The results did show substantial variation across slice thickness and kernels.
Conclusion: CAD nodule detection sensitivity was maintained across a wide range of dose levels and both reconstruction kernels used; however performance was substantially different across slice thicknesses.
Funding Support, Disclosures, and Conflict of Interest: Funding was provided in part by the University of California Office of the President Tobacco-Related Disease Research Program(UCOP-TRDRP grant #22RT-0131) and the National Cancer Institute of Quantitative Imaging Network(QIN grant U01-CA181156). Disclosure:The UCLA Department of Radiological Sciences has an Institutional Master Research Agreement with Siemens Healthineers(formerly Siemens Healthcare, Erlangen, Germany).
Contact Email: