Click here to


Are you sure ?

Yes, do it No, cancel

Photon-Counting Spectral CT with Principal Component Analysis in Image and Projection Domains

H Xie1, T Thuering2 , Y Liu3 , W Long1 , Y Ren1,3 , X Tang1*, (1) Emory University School of Medicine, Atlanta, GA, (2) DECTRIS Ltd., 5405 Baden-Daettwil, Switzerland, (3) SinoVision Technologies (Beijing) Co. Ltd., Beijing 100176, China


(Wednesday, 7/17/2019) 9:30 AM - 10:00 AM

Room: Exhibit Hall | Forum 9

Purpose: Photon-counting spectral CT has increasingly gained momentum in research and development, with the potential of overcoming the challenges imposed by advanced clinical applications, such as material decomposition and virtual monochromatic analysis. In this work, we investigate the feasibility and performance of principal component analysis (PCA) of extracting information from the data acquired with x-ray photon counting detector.

Methods: PCA is applied to the data acquired by x-ray photon counting detector in both image and projection domains, which transforms the data into an orthogonal coordinate system that maximizes its variances, and thus enhances the contrast between clustered data while reducing the dimensionality of data. The PCA in spectral CT may suffer from high noise in the high energy bins. A Content-Oriented Sparse Representation (COSR) de-noising method is carried out on the reconstructed images to investigate the benefit to PCA from COSR de-noising. Two phantoms (iodine; and iodine and nanoparticulated gold) and a lab mouse are used to evaluate the performance of PCA in photon counting spectral CT.

Results: It is found that the first principle component (PC) images extract more than 98.9%~99.8% of the information (covariance) carried by the images corresponding to all energy bins, in which the image quality (measured as noise and CNR) and material differentiation can be enhanced. By reducing noise in reconstructed images, the CNR in the first PC images can be enhanced by approximately 1.5-fold. It is also observed that the PCA can benefit from de-noising in images with enhanced ability of reducing data dimensionality. More studies of PCA in the projection domain are under the way.

Conclusion: According to the preliminary results, PCA showed its feasibility in extracting dominant contrasts between materials from and reducing the dimensionality of the photon counting spectral CT data.

Funding Support, Disclosures, and Conflict of Interest: X. Tang is recipient of a research grant from Sinovision Technologies (Beijing), Co. Ltd.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email