Iso-uncertainty control in an experimental fluoroscopy system

Authors

  • Siddique S.,

    1. Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada and Department of Computer Science, University of Toronto, Toronto, Ontario M5S 3G4, Canada
    Search for more papers by this author
  • Fiume E.,

    1. Department of Computer Science, University of Toronto, Toronto, Ontario M5S 3G4, Canada
    Search for more papers by this author
  • Jaffray D. A.

    1. Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada; Departments of Radiation Oncology, Medical Biophysics, and IBBME, University of Toronto, Toronto, Ontario M5S 3E2, Canada; and Techna Institute, Toronto, Ontario M5G 1P5, Canada
    Search for more papers by this author
    • a)

      Author to whom correspondence should be addressed. Electronic mail: david.jaffray@rmp.uhn.on.ca; Telephone: 416-946-4501 (x5387); Fax: 416-946-4578.


Abstract

Purpose:

X-ray fluoroscopy remains an important imaging modality in a number of image-guided procedures due to its real-time nature and excellent spatial detail. However, the radiation dose delivered raises concerns about its use particularly in lengthy treatment procedures (>0.5 h). The authors have previously presented an algorithm that employs feedback of geometric uncertainty to control dose while maintaining a desired targeting uncertainty during fluoroscopic tracking of fiducials. The method was tested using simulations of motion against controlled noise fields. In this paper, the authors embody the previously reported method in a physical prototype and present changes to the controller required to function in a practical setting.

Methods:

The metric for feedback used in this study is based on the trace of the covariance of the state of the system, tr(C). The state is defined here as the 2D location of a fiducial on a plane parallel to the detector. A relationship between this metric and the tube current is first developed empirically. This relationship is extended to create a manifold that incorporates a latent variable representing the estimated background attenuation. The manifold is then used within the controller to dynamically adjust the tube current and maintain a specified targeting uncertainty. To evaluate the performance of the proposed method, an acrylic sphere (1.6 mm in diameter) was tracked at tube currents ranging from 0.5 to 0.9 mA (0.033 s) at a fixed energy of 80 kVp. The images were acquired on a Varian Paxscan 4030A (2048 × 1536 pixels, ∼100 cm source-to-axis distance, ∼160 cm source-to-detector distance). The sphere was tracked using a particle filter under two background conditions: (1) uniform sheets of acrylic and (2) an acrylic wedge. The measured tr(C) was used in conjunction with a learned manifold to modulate the tube current in order to maintain a specified uncertainty as the sphere traversed regions of varying thickness corresponding to the acrylic sheets in the background.

Results:

With feedback engaged, the tracking error was found to correlate well with the specified targeting uncertainty. Tracking of the fiducial was found to be robust to changes in the attenuation presented by the varying background conditions. For a desired uncertainty of 5.0 mm, comparison of the feedback framework with a comparable system employing fixed exposure demonstrated dose savings of 29%.

Conclusions:

This work presents a relation between a state descriptor, tr(C), the x-ray tube current used, and an estimate of the background attenuation. This relation is leveraged to modulate the tube current in order to maintain a desired geometric uncertainty during fluoroscopy. The authors’ work demonstrates the use of the method in a real x-ray fluoroscopy system with physical motion against varying backgrounds. The method offers potential savings in imaging dose to patients and staff while maintaining tracking uncertainty during fluoroscopy-guided treatment procedures.

Ancillary