Removing Respiratory Artifacts in Nuclide Breast Imaging

Institution: Stanford University
Investigator(s): Brian Thorndyke, Ph.D. -
Award Cycle: 2005 (Cycle 11) Grant #: 11FB-0170 Award: $31,556
Award Type: Postdoctoral Fellowship
Research Priorities
Detection, Prognosis and Treatment>Imaging, Biomarkers, and Molecular Pathology: improving detection and diagnosis

Initial Award Abstract (2005)
Radionuclide imaging is a technique that allows physicians to obtain clear images of metabolically active tissues in the body. To obtain the images, radioactive materials (called “tracers”) are introduced into the patient’s body. The tracers emit gamma rays, which are the basis for detection. Radionuclide imaging holds great potential to non-invasively provide valuable information about breast cancer, including detection, staging, and treatment response. However, the utility of the imaging modality is heavily dependent on the sensitivity and accuracy of the reconstructed images. In breast scans, respiratory motion hinders tumor evaluation by effectively smearing the signal over the respiratory cycle, both lowering the signal-to-noise ratio and distorting the tumor volume.

A key issue in radionuclide imaging is the duration over which the scan must occur to obtain adequate statistics. While patient respiratory motion can be substantially controlled through various means in fast imaging modalities like computed tomography (CT), the several minutes required in PET scans severely hinders any possibility of respiratory control. Furthermore, any method which relies on partial signal acquisition (e.g., at end-inspiration) provides correspondingly less data per respiratory cycle, and consequently requires additional (and possibly unacceptable) scan time.

We hypothesize that respiratory motion artifacts in breast cancer can be eliminated through specialized acquisition and image processing methods, with minimal or no additional hardware beyond the radionuclide imaging equipment in current clinical use, and without requiring additional scanning time or patient effort. This project consists of three principal aims to achieve the goal of eliminating respiratory motion artifacts in breast cancer radionuclide imaging. The first aim is to construct a method of acquiring and sorting radionuclide data and respiratory motion information, optimally suited for breast cancer imaging. The second aim is to design and implement image processing algorithms that remove respiratory motion artifacts from the radionuclide images. The third aim is to validate these methods on breast cancer patient data. We propose to acquire in small time intervals, a fraction of the respiratory cycle, and then sort according to respiratory phase or amplitude. The sorted images can then be modified to match one another using deformable image registration methods, which permit tissue translation in addition to volume and shape distortion. Finally, stacking the deformed images will generate an image at a single respiratory point, without any loss of scan data. The method will be validated on approximately one dozen breast cancer patients enrolled in imaging and treatment protocols at the Stanford Hospital.

Because radionuclide scans are noninvasive, they are an attractive alternative to multiple biopsies in initial evaluation of tumor volume and stage. In addition, several treatment alternatives are readily available for breast cancer, and it would prove invaluable to develop reliable correlations between radionuclide scan information and treatment response at early stages of the treatment. Finally, accurate evaluation of the extent of the cancer, particularly for ductal carcinoma in situ, could make the difference between a lumpectomy and a radical mastectomy. Elimination of respiratory motion artifacts will permit the best possible realization of these potential achievements of radionuclide imaging.

Final Report (2006)
Respiratory motion can negatively affect the quality of positron emission tomography (PET) exams of the breast. Since PET exams typically require several minutes to acquire data, the resulting image represents an averaging of tumor motion over several breathing cycles. This has the effect of both blurring the image of the lesion, as well as decreasing its intensity. These artifacts can make it difficult to accurately delineate the target for surgery or radiation therapy, and can result in poorer diagnostic sensitivity in the first place. This project has focused on reducing imaging artifacts due to respiratory motion in breast cancer PET imaging.

The aims of the project were partially fulfilled. Methods were developed to acquire temporally divided PET data, and then recombine the data through deformable modeling into a single, enhanced image with reduced motion artifacts. These techniques were validated on phantom data, and developed to sufficient precision to aid, in principle, imaging artifacts during breast scans. The results demonstrated a significant enhancement of radionuclide signal over unprocessed scans, suggesting the potential to reveal small tumors that would otherwise be missed due to statistical noise. However, due to time limitations and early termination of the project, the methods were not applied prospectively to protocol breast patients. Furthermore, limitations with the PET/CT scanner inhibited complete manipulation of the raw data, which may have further enhanced the results. Nonetheless, the results of this study provide a firm basis for respiratory-artifact reduction in PET imaging of breast cancer.

While this project has been terminated in the context of this fellowship, the natural extension is to combine artifact-reduced breast PET imaging with artifact-reduced time-resolved CT imaging. So-called "4D CT" has been well-developed in radiation oncology over the past couple of years, although its PET counterpart has not. Fusion of artifact-reduced PET images with 4D CT may hold great benefits for diagnosis and treatment of breast cancer patients.