Breast tomosynthesis reconstruction with a multi-beam x-ray source

Ying Chen, Weihua Zhou, Guang Yang, X. Qian, Jianping Lü, Otto ZhouProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2009 · DOI

TL;DR

This paper introduces a novel multi-beam X-ray source for breast tomosynthesis, enabling faster 3D imaging with a stationary system. They evaluated five reconstruction algorithms, finding that while MITS and FBP offer good in-plane sharpness, BP, FBP, and MLEM excel at removing out-of-plane blur, especially with more projection data.

What Problem Does This Paper Address?

Traditional mammography provides a 2D projection of a 3D breast, leading to overlapping structures that can obscure abnormalities and hinder early cancer detection. Breast tomosynthesis aims to solve this by reconstructing 3D images from a series of low-dose X-ray projections. However, current tomosynthesis systems often use partial isocentric geometries where the X-ray tube moves in an arc, which can be uncomfortable for patients and introduce motion blur. This paper addresses the need for faster, more comfortable, and potentially higher-quality 3D breast imaging by exploring a novel stationary multi-beam X-ray source and evaluating its reconstruction performance with various algorithms.

Methods

The core of this work lies in the evaluation of a novel stationary multi-beam field-emission X-ray (MBFEX) source for breast tomosynthesis. This system utilizes carbon nanotube electron emitters and a parallel imaging configuration, where the X-ray sources are aligned in a plane parallel to the detector. This design allows for faster imaging speeds (e.g., 11.2 seconds for 25 views) compared to moving-tube systems, potentially reducing patient discomfort and motion blur.

The authors investigated five representative tomosynthesis reconstruction algorithms:

  • Back Projection (BP): A fundamental method for aligning projection data.
  • Filtered Back Projection (FBP): An extension of BP that applies a filter to improve image quality.
  • Matrix Inversion Tomosynthesis (MITS): A direct reconstruction method.
  • Maximum Likelihood Expectation Maximization (MLEM): An iterative algorithm that aims to maximize the likelihood of the observed data.
  • Simultaneous Algebraic Reconstruction Technique (SART): Another iterative algorithm.

Both phantom studies and computer simulations were used to evaluate these algorithms. The phantom study involved acquiring projection images of a sponge phantom with simulated masses. The computer simulation used a defined 3D spherical object and ray-tracing to generate datasets with varying numbers of projection images (9 and 25) and a fixed view angle (50 degrees).

Datasets

The paper utilizes two types of datasets:

  • Phantom Data: Projection images of a sponge phantom with embedded simulated masses were acquired using the novel multi-beam X-ray system. This dataset was used for a preliminary phantom study.
  • Simulated Data: Computer-generated tomosynthesis datasets of a 3D spherical object were created using a ray-tracing method. These datasets were generated with either 9 or 25 projection images and a total view angle of 50 degrees. This allowed for controlled evaluation of reconstruction algorithms under different projection numbers.

Key Insights and Ideas

  • Stationary Multi-Beam X-ray Source Potential: The novel stationary multi-beam X-ray source shows promise for faster and potentially more comfortable 3D breast imaging by eliminating the need for tube movement.
  • Algorithm Trade-offs: Different reconstruction algorithms exhibit distinct strengths and weaknesses. There isn't a single "best" algorithm for all aspects of tomosynthesis reconstruction.
  • In-Plane vs. Out-of-Plane Performance: Algorithms like MITS and FBP excel at enhancing in-plane structures, making them appear sharper. However, algorithms like BP, FBP, and MLEM are more effective at suppressing out-of-plane blur, which is crucial for isolating structures of interest.
  • Impact of Projection Number: Increasing the number of projection images significantly improves the quality of reconstruction, particularly in reducing out-of-plane artifacts and noise.

Key Results

  • Phantom Study: The reconstructed images from all five algorithms showed improved visibility of simulated masses compared to a single projection image, with clearer margins and shapes.
  • Computer Simulation (In-Plane):
    • BP, MLEM, and SART showed sharp responses in the in-plane reconstruction.
  • MITS and FBP demonstrated edge enhancement, which is also observed in current commercial systems.
  • Computer Simulation (Out-of-Plane):
    • With a limited number of projection images (9), all algorithms showed noise and artifacts in out-of-plane profiles.
  • With a larger number of projection images (25), BP, FBP, and MLEM performed better at removing out-of-plane blur, resulting in smoother profiles. FBP also showed edge enhancement in these profiles.
  • Speed: The multi-beam X-ray system is capable of generating 3D tomosynthesis images at a faster speed compared to current commercial prototype systems. Reconstruction times for BP, FBP, and MITS were under 3 minutes on a 2GHz CPU, while accelerated MLEM took about 2 minutes for three planes on a different system.

Limitations and Future Work

  • Preliminary Study: The authors emphasize that this was a preliminary phantom study and simulation.
  • Phantom Density: The dense nature of the phantom used in the study presented challenges for exposure level selection.
  • Calibration and Measurement Accuracy: Further work is needed to improve the calibration and measurement accuracy of imaging parameters.
  • Reconstruction Artifacts: Efforts are ongoing to reduce reconstruction artifacts.
  • Future Directions: The paper suggests that a larger number of projection images is beneficial for improved reconstruction quality. An improved design capable of generating 25 projection images has already been implemented. Future work will involve further phantom studies to optimize reconstruction and imaging geometry with the novel device.

Summary updated

Abstract

As a new three-dimensional breast imaging technique, breast tomosynthesis allows the reconstruction of an arbitrary set of planes in the breast from a limited-angle series of x-ray projection images. The breast tomosynthesis technique has been demonstrated as promising to improve early breast cancer detection. This paper represents a preliminary phantom study and computer simulation results of different breast tomosynthesis reconstruction algorithms with a novel carbon nanotube based multi-beam x-ray source. Five representative tomosynthesis reconstruction algorithms, including back projection (BP), filtered back projection (FBP), matrix inversion tomosynthesis (MITS), maximum likelihood expectation maximization (MLEM), and simultaneous algebraic reconstruction technique (SART) were investigated. Tomosynthesis projection images of a phantom were acquired with the stationary multi-beam x-ray tomosynthesis system. Reconstruction results from different algorithms were studied. A computer simulation study was further done to investigate the sharpness of reconstructed in-plane structures and to see how effective each algorithm is at removing out-of-plane blur with parallel-imaging geometries. Datasets with 9 and 25 projection images of a defined 3D spherical object were simulated with a total view angle of 50 degrees. Results showed that the multi-beam x-ray system is capable to generate 3D tomosynthesis images with faster speed compared with current commercial prototype systems. With simulated parallel-imaging geometry, MITS and FBP showed edge enhancement in-plane performance. BP, FBP and MLEM performed better at out-of-plane structure removal with larger number of projection images.
BibTeX
@inproceedings{chen2009breast,
  author = {Ying Chen and Weihua Zhou and Guang Yang and X. Qian and Jianping Lü and Otto Zhou},
  title = {{Breast tomosynthesis reconstruction with a multi-beam x-ray source}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2009},
  doi = {10.1117/12.811659},
}
Figure