Related Experiment Video
Updated: Jul 20, 2025
10:24
Neutron Radiography and Computed Tomography of Biological Systems at the Oak Ridge National Laboratory's High Flux Isotope Reactor
Published on: May 7, 2021
2.4K
Constrained one-step material decomposition reconstruction of head CT data from a silicon photon-counting prototype
Taly Gilat Schmidt1, Emil Y Sidky2, Xiaochuan Pan2
1Department of Biomedical Engineering, Marquette University and Medical College of Wisconsin, Milwaukee, Wisconsin, USA.
Medical Physics
|July 31, 2023
Summary
The constrained One-Step Spectral CT Image Reconstruction (cOSSCIR) algorithm offers a more stable method for material decomposition in photon-counting CT (PCCT) imaging. This advanced technique significantly reduces noise and improves image quality in clinical head CT scans.
Area of Science:
- Medical Imaging
- Quantitative Imaging
- Photon-Counting CT
Background:
- Spectral CT material decomposition is crucial for quantitative imaging but faces challenges with inversion stability.
- The constrained One-Step Spectral CT Image Reconstruction (cOSSCIR) algorithm was developed to stabilize this inversion process by directly estimating basis material images.
Purpose of the Study:
- To evaluate the performance of the cOSSCIR algorithm on clinical head CT datasets from a photon-counting CT (PCCT) prototype.
- This study represents the first investigation of cOSSCIR on large-scale, anatomically complex clinical PCCT data.
- The cOSSCIR method incorporates spectrum estimation and nonlinear counts correction for nonideal detector effects.
Main Methods:
- Head CT data were acquired using a clinical PCCT prototype with an eight-energy-bin silicon detector.
- Calibration data were used to train spectral and nonlinear counts correction models for detector effects.
- The cOSSCIR algorithm directly optimized bone and adipose basis images with a grouped total variation (TV) constraint, compared against two-step Maximum Likelihood Estimation (MLE) followed by filtered backprojection (FBP) or TV minimization (MLE + TVmin).
Main Results:
- cOSSCIR reduced noise standard deviation in basis images by 2-6x compared to MLE + TVmin at equivalent TV constraints.
- cOSSCIR images showed improved spatial resolution and fine anatomical detail.
- While MLE + TVmin yielded lower noise for higher-energy virtual monoenergetic images (VMIs), cOSSCIR VMIs had lower noise at lower energies and better qualitative resolution. cOSSCIR soft-tissue mean values were closer to expected values.
Conclusions:
- The cOSSCIR algorithm, coupled with spectral modeling and nonlinear counts correction, successfully reconstructed bone and adipose basis images from clinical PCCT data.
- cOSSCIR demonstrated superior noise reduction (2-6x) and enhanced depiction of fine anatomical details compared to the MLE + TVmin two-step approach.
