TU-H-206-01: An Automated Approach for Identifying Geometric Distortions in Gamma Cameras

Authors

  • Mann S,

    1. Clinical Imaging Physics Group and Carl E. Ravin Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, North Carolina 27705 and Medical Physics Graduate Program, Duke University, Durham, North Carolina 27705
    Search for more papers by this author
  • Nelson J,

    1. Clinical Imaging Physics Group and Carl E. Ravin Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, North Carolina 27705 and Medical Physics Graduate Program, Duke University, Durham, North Carolina 27705
    Search for more papers by this author
  • Samei E

    1. Clinical Imaging Physics Group and Carl E. Ravin Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, North Carolina 27705 and Departments of Physics, Electrical and Computer Engineering, and Biomedical Engineering, and Medical Physics Graduate Program, Duke University, Durham, North Carolina 27705
    Search for more papers by this author

Abstract

Purpose:

To develop a clinically-deployable, automated process for detecting artifacts in routine nuclear medicine (NM) quality assurance (QA) bar phantom images.

Methods:

An artifact detection algorithm was created to analyze bar phantom images as part of an ongoing QA program. A low noise, high resolution reference image was acquired from an x-ray of the bar phantom with a Philips Digital Diagnost system utilizing image stitching. NM bar images, acquired for 5 million counts over a 512×512 matrix, were registered to the template image by maximizing mutual information (MI). The MI index was used as an initial test for artifacts; low values indicate an overall presence of distortions regardless of their spatial location. Images with low MI scores were further analyzed for bar linearity, periodicity, alignment, and compression to locate differences with respect to the template. Findings from each test were spatially correlated and locations failing multiple tests were flagged as potential artifacts requiring additional visual analysis. The algorithm was initially deployed for GE Discovery 670 and Infinia Hawkeye gamma cameras.

Results:

The algorithm successfully identified clinically relevant artifacts from both systems previously unnoticed by technologists performing the QA. Average MI indices for artifact-free images are 0.55. Images with MI indices < 0.50 have shown 100% sensitivity and specificity for artifact detection when compared with a thorough visual analysis. Correlation of geometric tests confirms the ability to spatially locate the most likely image regions containing an artifact regardless of initial phantom orientation.

Conclusion:

The algorithm shows the potential to detect gamma camera artifacts that may be missed by routine technologist inspections. Detection and subsequent correction of artifacts ensures maximum image quality and may help to identify failing hardware before it impacts clinical workflow. Going forward, the algorithm is being deployed to monitor data from all gamma cameras within our health system.

Ancillary