The chest radiograph is a powerful tool for diagnosing many diseases, but is suboptimal for detection of pulmonary nodules. It has been estimated that 30% of pulmonary nodules are missed on initial reading of chest radiographs. The main limitation for detecting these nodules is not quantum noise or even low contrast but rather poor conspicuity. The nodules are hard to find against the background of overlying anatomy.

One method to reduce the visual clutter from overlying anatomy is dual-energy imaging. As the name implies, dual-energy radiography involves taking two radiographs at different mean beam energies (Fig 1a, 1b). These radiographs are then combined to form a subtraction image that highlights either the soft-tissue or bone components. The soft-tissue image (Fig 1c) shows better visualization of pulmonary nodules because the ribs are made to vanish. The bone image (Fig 1d) shows if a given nodule is calcified.

Figure 1. Dual-energy radiography. (a: top left) conventional low-energy radiograph. (b: top right) conventional high-energy radiography. (c: bottom left) Dual-energy soft-tissue image. Note the excellent suppression of the ribs. (d: bottom right) Dual-energy bone image. (See Dobbins and Warp Advances in Digital Radiography Samei E, Flynn MJ (eds), RSNA, Oak Brook, Illinois, 2003.)

The difference in energy between the two radiographs can be accomplished in two ways. One method is to use a sandwich detector comprised of two computed radiography (CR) plates with an interplate filter to harden the beam for the image in the second plate. This is the method used with commercial computed radiography systems that offer dual-energy imaging. An alternative approach is to use two separate image acquisitions, one at a high kV and one at a low kV. This is the method used with flat-panel detectors. The two-exposure method has far better signal-to-noise ratio, but is subject to some slight motion misregistration due to patient breathing or cardiac motion during the short time interval between the two image acquisitions.

Our laboratory has over 25 years of experience in developing, optimizing, and evaluating dual-energy imaging for pulmonary nodule detection. This work includes 2 NIH grants (R01 CA55388 and continuation) and research agreements with General Electric. General Electric offers a commercial dual-energy product based partly on work done in collaboration with our laboratory.

We have developed algorithms for use with both CR and flat-panel detectors. As examples of the types of algorithms we have developed, we have devised scatter models to correct for scattered radiation within the sandwich-type CR detector and from the patient. We also have developed a calibrated basis-material decomposition technique that more accurately determines bone and tissue thickness than the typical energy subtraction method. Fig. 2 shows bone and soft-tissue images generated with our self-calibrating basis material method. Note the excellent delineation of soft-tissue and bone. There are, however, some misregistration artifacts seen in the bone image due to the two-image acquisition approach; these artifacts can potentially be minimized in the future using ECG gating methods that we are investigating.

Figure 2. Dual-energy chest radiography. (a: left) Dual-energy soft-tissue image. Note the excellent suppression of the ribs. (b: right) Dual-energy bone image. There are slight misregistration artifacts in the bone image from the two-image method. These images were produced from human subject images collected by GE Healthcare in collaboration with our laboratory, using the self-calibrating basis material tissue/bone decomposition method we developed.

Figure 3. Noise power spectrum comparison of various dual-energy noise suppression methods. Noise suppression methods evaluated included noise clipping (NOC), correlated noise reduction (KCNR), spatial frequency filtering of the high-energy image (SSF on H), edge predictive adaptive smoothing (EPAS), and combinations of these methods. The solid line at the top is the NPS for an uncorrected dual-energy image. These curves represent NPS of the bone image in the subdiaphragm region. (See Dobbins and Warp Advances in Digital Radiography Samei E, Flynn MJ (eds), RSNA, Oak Brook, Illinois, 2003.)

Our lab has also worked extensively to develop noise-suppression methods for dual-energy imaging, and to quantitatively compare various such methods. Figure 3 shows the noise power spectrum of several noise-suppression methods, indicating that while low-frequency noise remains a challenge, mid-to-high frequency noise can be successfully suppressed by over a factor of 10 using combinations of various noise suppression methods.

More recently, we have conducted an NIH-funded clinical trial to investigate the relative performance of dual-energy, tomosynthesis, and conventional PA/lateral chest radiography for pulmonary nodule detection. The observer study for that trial is still underway.

Representative publications

• Hinshaw DA, Dobbins JT III: Recent progress in noise reduction and scatter correction in dual-energy imaging. Proc. SPIE Medical Imaging Symposium, 2432:134-142, 1995.
• Hinshaw DA, Dobbins JT III: Plate scatter correction for improved performance in dual-energy chest imaging. Medical Physics 23:871-876, 1996.
• Warp RJ, Dobbins JT III: Quantitative evaluation of noise reduction strategies in dual-energy imaging. Medical Physics 30(2):190-198, 2003.
• Dobbins JT III, Warp RJ: Dual-energy methods for tissue discrimination in chest radiography. In Advances in Digital Radiography: RSNA Categorical course in Diagnostic Radiology Physics, Samei E, Flynn MJ (eds), Radiological Society of North America, Oak Brook, Illinois, 2003.
  

Acknowledgments
Grant support for this project was provided by the National Institutes of Health (R01 CA55388) and GE Healthcare. Duke University and GE Healthcare jointly are inventors on a patent on ECG-gating in dual-energy imaging.