Quantitative Assessment of Segmentation Accuracy in Single-Source and Multisource Cone-Beam CT

Abstract

Objective: Conventional single-source cone-beam CT (CBCT) is affected by x-ray scatter and cone-beam artifacts that limit quantitative structural measurements. This study compared single-source CBCT with a multisource CBCT (ms-CBCT) architecture based on a carbon nanotube x-ray source array, using clinical multidetector CT (MDCT) as the reference. Methods: A head phantom was imaged on a switchable benchtop system in both configurations at 90 and 110 kVp. Volumes were reconstructed with a simultaneous iterative reconstruction technique and total-variation denoising, registered to a common frame, and resampled to the MDCT grid. Bone was segmented using Hounsfield-unit thresholds, and agreement with MDCT was assessed by surface area, segmented volume, and Dice similarity coefficient (DSC). Results: At 90 kVp, ms-CBCT reduced surface-area error from 5.8% to 2.6% and volume error from 15.9% to 7.1%, and raised DSC from 0.74 to 0.80. At 110 kVp, it reduced surface-area error from 14.0% to 3.4% and volume error from 19.7% to 10.1%, and raised DSC from 0.74 to 0.82. Conclusion: At both tube voltages, ms-CBCT agreed more closely with MDCT than the single-source configuration. Using surface area, volume, and DSC together gives a more complete assessment of quantitative CBCT performance.