Figure 1. Importance of deblurring in tomosynthesis.
	(a) Conventional shift-and-add tomosynthesis reconstuction with no deblurring. (b) Deblurred matrix inversion tomosynthesis image of same plane showing greatly improved slice definition.

Chest tomosynthesis has shown the potential to greatly improve detection of pulmonary nodules. A recent publication from our group showed that chest tomosynthesis using flat-panel detectors can triple the detection sensitivity of subtle pulmonary nodules compared with conventional chest radiography. (see also link to page on clinical evaluations of chest tomosynthesis).

Our laboratory has been involved in chest tomosynthesis research for over 20 years (including 2 NIH grants [R01 CA80490 and continuation] and 2 corporate research agreements), and we will describe here some of the experimental work done to optimize the technique for pulmonary nodule detection.

Part of our work in chest tomosynthesis has focused on developing improved algorithms for reconstruction. Our reconstruction algorithm, Matrix Inversion Tomosynthesis (MITS), uses linear algebra to solve for the blur that occurs from anatomy above and below the plane of interest in a tomosynthesis reconstruction. In order to get high quality slice images, it is imperative to use a deblurring algorithm (Fig 1). MITS performs very well at removing residual tomosynthesis image blur.

In order to optimize the image acquisition process for a particular detection task with a given reconstruction algorithm, one must determine the best total tube angle, the optimum number of projection images, and the best plane spacing for reconstructed images. We constructed the   world’s  first   flat-panel-based  tomosynthesis  device

using a  commercial-grade  detector and a customized tube

Figure 2. Prototype experimental chest tomosynthesis apparatus constructed in our laboratory. Depicted is the computer-controlled custom-built x-ray tube mover.

mover (Fig 2), and conducted experiments to determine the optimum acquisition parameters. Simulated impulse response functions (and MTFs) were produced to evaluate MITS compared to conventional shift-and-add tomosynthesis for a variety of acquisition parameters. Figure 3 shows the three-dimensional impulse response for MITS and conventional shift-and-add tomosynthesis for 11 and 61 projection images. As can be seen, 61 projection images give a far smoother impulse response function. We also measured noise power spectra (NPS) of reconstructed tomosynthesis images, and found that the NPS was very consistent between MITS and shift-and-add tomosynthesis except in the mid-range frequencies. The NPS curves (Fig 4) show how the reconstructed image quality varies with different acquisition parameters.

Based on these optimization experiments, we determined that the best acquisition parameters for detection of pulmonary nodules using the MITS algorithm are 71 projection images, 20-degrees of total tube movement, and 5 mm plane spacing. Acquisition parameters optimized for other reconstruction algorithms such as filtered backprojection are slightly different.

See the review article written by our laboratory (PMB, 2003, cited below) for a general overview of tomosynthesis methods and clinical applications.

Conv Tomo
MITS
	11 projection images	61 projection images
Figure 3. Impulse response for chest tomosynthesis. Shown are examples for both MITS and conventional shift-and-add (SAA) tomosynthesis. (a) SAA with 11 projection images. (b) SAA with 61 projection images. (c) MITS with 11 projection images. (d) MITS with 61 projection images. Note the improved uniformity of response with 61 projection images compared with 11 projection images. MITS provides better suppression of out-of-plane structures.

Exposure x Normalized NPS
Anterior Plane		Central Plan
20-degree tube motion, 19 planes, 16.7 mm plane spacing
Figure 4. Noise power spectra of reconstructed tomosynthesis images. The NPS was measured using a uniform acrylic phantom and was normalized by incident exposure (ENNPS). ENNPS is shown as a function of number of projection images for (a) an anterior plane and (b) a central plane in the phantom. Note that the main variation with projection image number occurs in the midrange frequencies at less than 61 projection images.

Representative publications

• Godfrey DJ, Warp RJ, Dobbins JT III: Optimization of Matrix Inversion Tomosynthesis. Proc. SPIE Medical Imaging 2001 Symposium, 4320:696-704, 2001.
• Godfrey DJ, Rader A, Dobbins JT III: Practical strategies for the clinical implementation of matrix inversion tomosynthesis. Proc. SPIE Medical Imaging 2003 Symposium, 5030:379-390, 2003.
• J. T. Dobbins, III and D. J. Godfrey, "Digital x-ray tomosynthesis: current state of the art and clinical potential," Phys. Med. Biol. 48, R65-R106 (2003).
• Godfrey DJ, McAdams HP, Dobbins JT III: Optimization of the matrix inversion tomosynthesis (MITS) impulse response and modulation transfer function characteristics for chest imaging. Medical Physics 33(3):655-667, 2006.
• Dobbins JT III, McAdams HP, Song J-W, Li CM, Godfrey DJ, DeLong DM, Paik S-H, Martinez-Jimenez S: Digital tomosynthesis of the chest for lung nodule detection: interim sensitivity results from an ongoing NIH-sponsored trial. Medical Physics 35(6):2554-2557, 2008.
• Godfrey DJ, McAdams HP, Dobbins JT III: Stochastic noise characteristics in matrix inversion tomosynthesis (MITS). Medical Physics 36(5):1521-1532, 2009.
• Dobbins JT III: Tomosynthesis imaging: at a translational crossroads. Medical Physics 36(6):1956-1967, 2009.
• Dobbins JT III, McAdams HP: Chest tomosynthesis: technical principles and clinical update. European J Radiology (in press), 2009.

Acknowledgments
Grant support for this project was provided by the National Institutes of Health (R01 CA080490) and GE Healthcare. Duke University and GE Healthcare jointly hold a patent on tube movement strategy in tomosynthesis.