Extracting respiratory signals from thoracic cone beam CT projections

Hao Yan, Xiaoyu Wang, Wotao Yin, Tinsu Pan, Moiz Ahmad, Xuanqin Mou, Laura Cerviño, Xun Jia, Steve JiangPhysics 2013 · DOI · arXiv

Abstract

The patient respiratory signal associated with the cone beam CT (CBCT) projections is important for lung cancer radiotherapy. In contrast to monitoring an external surrogate of respiration, such a signal can be extracted directly from the CBCT projections. In this paper, we propose a novel local principal component analysis (LPCA) method to extract the respiratory signal by distinguishing the respiration motion-induced content change from the gantry rotation-induced content change in the CBCT projections. The LPCA method is evaluated by comparing with three state-of-the-art projection-based methods, namely the Amsterdam Shroud method, the intensity analysis method and the Fourier-transform-based phase analysis method. The clinical CBCT projection data of eight patients, acquired under various clinical scenarios, were used to investigate the performance of each method. We found that the proposed LPCA method has demonstrated the best overall performance for cases tested and thus is a promising technique for extracting a respiratory signal. We also identified the applicability of each existing method.

Introduction

In lung cancer radiotherapy, precision is paramount. Because tumors move as a patient breathes, clinicians utilize 4D Cone Beam CT (CBCT) to map this motion and target radiation accurately. Traditionally, external surrogates—such as pressure belts or skin markers—track breathing. However, external movement does not always perfectly correlate with internal lung anatomy.

A superior alternative is extracting the breathing signal directly from X-ray projections. The primary challenge lies in the signal being a complex mixture of two periodic motions: the patient’s respiration and the 360-degree rotation of the CT gantry. Separating the subtle respiratory "wobble" from the massive "sweep" of the scanner rotation is a significant signal-processing hurdle, particularly when high-contrast landmarks like the diaphragm are absent from the field of view.

Methods

The study compares three established projection-based methods with a novel approach called Local Principal Component Analysis (LPCA).

Comparison Methods

The three state-of-the-art methods evaluated are:

  1. Amsterdam Shroud (AS): Uses line-by-line registration to calculate pixel shifts.
  2. Intensity Analysis (IA): Extracts signals based on intensity summation across views.
  3. Fourier-Transform Phase (FT-p): Extracts phase information from the 2D Fourier transform of the projections.
Overview of the three state-of-the-art projection-based methods: (a) Amsterdam Shroud, (b) Intensity Analysis, and (c) Fourier-transform phase analysis.

Local Principal Component Analysis (LPCA)

The proposed LPCA method is built on a geometric insight: if each X-ray projection is represented as a point in high-dimensional space, the entire scan forms a 2-torus manifold. One circular path represents gantry rotation, while the other represents the respiratory cycle.

The LPCA pipeline involves:

  • Preprocessing: Generating an Amsterdam Shroud (AS) image to highlight motion.
  • Foreground Extraction: Using a Total Variation (TV/L1) model to isolate breathing patterns from the smooth background.
  • Local PCA: Applying PCA within a sliding window to "unroll" the torus locally, finding the plane where breathing motion is most distinct.
Geometric representation of the respiratory signal extraction: (a) gantry rotation, (b) the breathing circle, and (c) the local PCA approach using different window sizes (W) to separate the motions.

Datasets

The methods were tested on clinical CBCT data from eight patients, encompassing various challenges including different scan speeds (1-minute vs 4-minute), field of view limitations (with/without diaphragm), and dose levels. Results were validated against commercial Real-time Position Management (RPM) signals and manual benchmarks.

Results and Discussion

LPCA demonstrated superior accuracy and robustness across all patient cases. Key findings include:

  • Landmark Independence: Unlike the AS method, which requires sharp moving edges like the diaphragm, LPCA extracts signals from the collective variance of the entire image.
  • ROI Robustness: While the FT-p method is highly sensitive to the Region of Interest (ROI), LPCA provides consistent results regardless of the detector area analyzed.
  • Clinical Impact: In one case, the LPCA-derived signal improved internal anatomy placement by 4.4 mm compared to Intensity Analysis, a critical margin for precision radiotherapy.
  • Efficiency: Processing averaged 72.8 ms per frame, indicating potential for near real-time clinical application.

Conclusion

The LPCA method effectively distinguishes respiratory motion from gantry rotation by leveraging the underlying manifold geometry of CBCT projections. While all methods face challenges in "half-fan" detector modes, LPCA's overall performance makes it a promising tool for both retrospective 4D-CT reconstruction and future real-time tumor tracking.

Summary updated

Abstract

The patient respiratory signal associated with the cone beam CT (CBCT) projections is important for lung cancer radiotherapy. In contrast to monitoring an external surrogate of respiration, such a signal can be extracted directly from the CBCT projections. In this paper, we propose a novel local principal component analysis (LPCA) method to extract the respiratory signal by distinguishing the respiration motion-induced content change from the gantry rotation-induced content change in the CBCT projections. The LPCA method is evaluated by comparing with three state-of-the-art projection-based methods, namely the Amsterdam Shroud method, the intensity analysis method and the Fourier-transform-based phase analysis method. The clinical CBCT projection data of eight patients, acquired under various clinical scenarios, were used to investigate the performance of each method. We found that the proposed LPCA method has demonstrated the best overall performance for cases tested and thus is a promising technique for extracting a respiratory signal. We also identified the applicability of each existing method.
BibTeX
@inproceedings{yan2013extracting,
  author = {Hao Yan and Xiaoyu Wang and Wotao Yin and Tinsu Pan and Moiz Ahmad and Xuanqin Mou and Laura Cerviño and Xun Jia and Steve Jiang},
  title = {{Extracting respiratory signals from thoracic cone beam CT projections}},
  booktitle = {Physics},
  year = {2013},
  doi = {10.1088/0031-9155/58/5/1447},
}
Figure