Point spread function modeling and image restoration for cone-beam CT

Hua Zhang, Kuidong Huang, Yikai Shi, Zhe XuChinese 2015 · DOI · arXiv

Abstract

X-ray cone-beam computed tomography (CT) offers high efficiency and precision for medical and industrial imaging. However, inherent degradation—primarily blur—limits image quality. This paper presents a novel point spread function (PSF) modeling method that establishes a general model for cone-beam CT. By mapping PSF parameters to scanning conditions (voltage and brightness), the model enables direct projection restoration without repeated physical measurements. Additionally, an enhanced restoration algorithm utilizing pre-filtering and pre-segmentation is introduced to sharpen edges while maintaining natural noise levels. Experimental results confirm the effectiveness of these methods in improving CT image clarity.

The Challenge of Image Degradation in CBCT

In cone-beam CT (CBCT), images are often blurred by two physical phenomena: the finite size of the X-ray source's focal spot and light scattering within the flat panel detector (FPD). Mathematically, this blur is represented by a Point Spread Function (PSF).

Schematic of the cone-beam CT imaging system used for PSF measurement.

Traditional restoration requires measuring the PSF for every unique set of scanning parameters (e.g., voltage, dose) using specialized targets like pinholes or gratings. This process is time-consuming and impractical for dynamic industrial environments. Furthermore, standard deblurring algorithms often amplify noise or introduce "ringing" artifacts, which can obscure critical details.

General PSF Modeling

To overcome the need for constant recalibration, researchers developed a "universal" PSF model. By measuring the system's response at 16 specific combinations of X-ray voltage (100–400 kV) and brightness (1000–2500 gray levels), they observed that the blur follows a predictable 3D Gaussian distribution.

A pinhole projection image (a) and its corresponding 3D Gaussian PSF model (b).

The parameters of this Gaussian model ($a$ and $b$) were found to vary deterministically with the scanning conditions. By fitting these variations to high-order polynomials, the authors created a continuous model capable of calculating the PSF for any arbitrary condition within the tested range. This allows for "on-the-fly" restoration without additional hardware measurements.

Enhanced Restoration Algorithm

The proposed restoration approach improves upon the standard Lucy-Richardson (LR) algorithm through two key innovations:

  1. Noise Management via Pre-filtering: Deblurring typically amplifies high-frequency noise. To prevent this, the authors use a Non-Local Means (NLM) algorithm to separate the image into a "clean" signal and a "noise" component. Only the clean signal is deblurred; the original noise is then added back, preserving the natural texture of the image while sharpening the edges.
  2. Efficiency via Pre-segmentation: Iterative deblurring is computationally expensive. By identifying the object's bounding box and ignoring the empty background, the algorithm significantly reduces processing time, making it feasible for large batches of projection images.

Key Results and Discussion

The methodology was validated using an aluminum industrial part. Compared to standard techniques like Wiener filtering and Constrained Least Squares (CLS), the proposed method achieved:

  • Superior Sharpness: The Average Gradient (AG) increased from 30.62 in the original image to 36.06, indicating significantly clearer edges.
  • Stable Noise Levels: Unlike other filters that degraded the Signal-to-Noise Ratio (SNR), the proposed method maintained an SNR (61.63) nearly identical to the original (62.35).
  • Artifact Suppression: Visual analysis showed a complete absence of the "ringing" effects typically seen in frequency-domain restoration.

Conclusion

The proposed general PSF model and enhanced restoration algorithm provide a practical solution for high-quality CBCT imaging. By automating PSF calculation and intelligently managing noise, the system achieves sharper, more accurate reconstructions suitable for demanding non-destructive testing and medical applications.

Summary updated

Abstract

X-ray cone-beam computed tomography (CT) has the notable features such as high efficiency and precision, and is widely used in the fields of medical imaging and industrial non-destructive testing, but the inherent imaging degradation reduces the quality of CT images. Aimed at the problems of projection images degradation and restoration in cone-beam CT, a point spread function (PSF) modeling method is proposed firstly. The general PSF model of cone-beam CT is established, and based on it, the PSF under arbitrary scanning conditions can be calculated directly for projection images restoration without the additional measurement, which greatly improved the application convenience of cone-beam CT. Secondly, a projection images restoration algorithm based on pre-filtering and pre-segmentation is proposed, which can make the edge contours in projection images and slice images clearer after restoration, and control the noise in the equivalent level to the original images. Finally, the experiments verified the feasibility and effectiveness of the proposed methods.
BibTeX
@inproceedings{zhang2015point,
  author = {Hua Zhang and Kuidong Huang and Yikai Shi and Zhe Xu},
  title = {{Point spread function modeling and image restoration for cone-beam CT}},
  booktitle = {Chinese},
  year = {2015},
  doi = {10.1088/1674-1137/39/3/038202},
}
Figure