Premium
Evaluation of an automated grid artifact detection system for quality control in digital mammography
Author(s) -
MacLellan Christopher J.,
Layman Rick R.,
Geiser William,
Gress Dustin A.,
Jones A. Kyle
Publication year - 2019
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1002/mp.13621
Subject(s) - imaging phantom , mammography , grid , digital mammography , computer science , image quality , computer vision , artifact (error) , artificial intelligence , medical imaging , medical physics , nuclear medicine , medicine , image (mathematics) , mathematics , geometry , cancer , breast cancer
Purpose Grid artifacts occur in digital mammography when synchronization between the grid assembly and generator is not achieved, including when malfunctions occur in the grid assembly or generator subsystems. Such artifacts are not explicitly monitored or evaluated by existing mammography quality control programs. In this study, we developed an automated method for quantifying the presence of grid artifacts in two‐dimensional (2D) digital mammography images and assessed its utility as a supplement to existing quality control programs. Methods Four digital mammography systems (Hologic Dimensions 3D 5000) were configured to automatically transfer 2D images to a server where the strength of the grid pattern, γ max , was quantified using a template‐matching algorithm and stored in amySQL database. This analysis was performed on both American College of Radiology (ACR) phantom and clinical images. Changes in γ max were compared with image quality and service records to establish preliminary action limits for physicist intervention for each type of image. These action limits were applied around selected service events to evaluate their clinical utility. Results All systems exhibited a gradual increase in γ max in ACR phantom images prior to having identical major components of the generator subsystem replaced, despite the absence of visible gridlines in the images. Retrospective analysis of phantom images suggested that physicists should consider AEC testing when γ maxexceeds 0.050 and that clinical image quality may be affected when γ maxexceeds 0.060. Eighteen of 19 visible grid artifacts were identified using a threshold γ maxvalue of 0.065 in clinical images. Warning limits that indicate abnormal operation before visible degradation in image quality were also established. These warning limits were 0.046 and 0.041 for the 24 × 29 cm and 18 × 24 cm paddles, respectively. Specific malfunctions in the generator and grid subsystems can be detected by applying these limits. Conclusions Automated monitoring of γ maxprovides useful information about the status of digital mammography units without affecting clinical operations. When used with appropriate action limits, this type of monitoring can help physicists identify specific equipment malfunctions before they would be detected by other quality control tests and before they affect clinical images.