(Monday, 7/30/2018) 7:30 AM - 9:30 AM
Room: Karl Dean Ballroom A1
Purpose: Material decomposition is the basis of spectral CT applications. The conventional strategy for generating material-specific images is based on matrix inversion or least-square fitting which is prone to noise amplification. Here, we describe a Material Decomposition framework using Spectral Prior Image Constrained Compressed Sensing (MD-SPICCS) to reduce the noise in the material-specific images while preserving image details and spatial resolution.
Methods: The proposed MD-SPICCS exploits the structural redundancy between material-specific images and spectral CT images by incorporating a low-noise prior image into its objective function. The use of low-noise prior serves to retain image details in material-specific image while achieving effective denoising. To test the proposed method, a multi-energy CT phantom with iodine and hydroxyapatite (HA) inserts of different concentrations and diameters were scanned on a whole-body Photon-Counting-Detector (PCD)-CT system. Iodine/HA-based decomposition was performed using standard least-square-based material decomposition (MD-LS), total-variation-regularized decomposition (MD-TV), and the proposed MD-SPICCS. The root-mean-square-errors (RMSE) of iodine/HA quantification were measured. A swine model was further scanned on PCD-CT system; iodine/fat-based decomposition was performed. Finally, in vivo human brain images was acquired on a dual-source dual-energy CT, and used for iodine/HA-based decomposition.
Results: The material-specific images generated from phantom data using MD-LS suffer from strong noise contamination (iodine/HA RMSE=3.9/80.8mg/mL). The MD-TV reduces the noise level substantially (iodine/HA RMSE=1.0/20.0mg/mL) but also loses image details. The MD-SPICCS can achieve effective noise reduction (iodine/HA RMSE=1.0/20.5mg/mL), while better preserving the low-concentration or small-diameter inserts. These observations are further confirmed by in vivo animal and human experiment data, which demonstrate that MD-SPICCS can improve delineation of small vasculatures in iodine-specific images while reducing image noise, and better preserve structural texture and image details in HA/fat-specific images.
Conclusion: We developed a novel material decomposition framework which reduced RMSE of iodine quantification by 75% while preserving image details in the material-specific images.
Funding Support, Disclosures, and Conflict of Interest: CHM receives industry grant support from Siemens Healthcare.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
