Abstract Archives of the RSNA, 2010
SSA21-05
Automatic Image Registration of Three-dimensional Ultrasound Thyroid Images: Application for Longitudinal Thyroid Nodule Tracking
Scientific Formal (Paper) Presentations
Presented on November 28, 2010
Presented as part of SSA21: Physics (Ultrasound)
Derek William Cool PhD, Presenter: Nothing to Disclose
Aaron D Ward PhD, Abstract Co-Author: Nothing to Disclose
Barry Brandon Hobbs MD, Abstract Co-Author: Nothing to Disclose
Walter Matthew Romano MD, Abstract Co-Author: Nothing to Disclose
Merrill Edmonds MD, Abstract Co-Author: Nothing to Disclose
Aaron Fenster PhD, Abstract Co-Author: Nothing to Disclose
Longitudinal, cancer-risk assessment is complex in multi-nodular glands because identification of corresponding nodules can be difficult using 2D US. 3D US thyroid images contain the complete gland and allow for image registration to automatically correspond nodules from multiple 3D scans. The accuracy of 3D image registration was evaluated on thyroid patients.
5 thyroid patients were selected based on a clinical assessment of cancer risk (ethics board approved). Multiple 3D US scans were acquired for each thyroid lobe. 3D US images were acquired using a 12 MHz transducer with a linear translator developed in our laboratory. For each 3D US scan, the probe was removed and repositioned on the neck to ensure a unique location for each image volume. A total of 16 3D US scans from 8 thyroid lobes were collected (2 patients had single lobes). Intra-lobe rigid registration of each patient’s thyroid gland, was achieved using a fully-automatic, block-matching, mutual-information based, 3D image registration algorithm. Anatomical fiducials were identified for each thyroid lobe pair and used to evaluate the 3D image registration accuracy. The targeted registration accuracy was <5mm or the minimum radius of a nodule that may be considered for biopsy.
16 nodules were identified within the thyroids, with a range of diameters of 3.5-25mm. Prior to registration, only 38% (6/16) of nodules were partially overlapped between corresponding 3D US scans and the mean fiducial distance error was 9.1±9.5mm. Following image registration, 88% (14/16) of nodules overlapped and the mean fiducial error was 4.3±4.0mm. One thyroid lobe registration was a large outlier as the two 3D US thyroid scans were substantially misaligned pre-registration (RMS=31.5mm), and were only partially corrected post-registration (RMS=13.7mm). The mean fiducial error for pre- and post-registration was 5.9±3.2mm and 3.0±1.5mm, respectively, when the outlier was excluded. This post-registration error was significantly less than the 5mm target representing the smallest nodule radius considered for biopsy (p<0.05).
Image registration can automatically and accurately provide correspondence of thyroid nodules seen on multiple 3D US scans.
3D US thyroid imaging with automatic image registration can improve thyroid nodule correspondence across multiple scans and might enhance longitudinal screening for thyroid malignancy.
Cool, D,
Ward, A,
Hobbs, B,
Romano, W,
Edmonds, M,
Fenster, A,
Automatic Image Registration of Three-dimensional Ultrasound Thyroid Images: Application for Longitudinal Thyroid Nodule Tracking . Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL.
http://archive.rsna.org/2010/9012804.html