Premium
SU‐C‐134‐01: CT Performance Assessment Using Statistical Processing Control Cloud‐Based Image Processing
Author(s) -
Fredriksson J,
Olafsdottir H,
Levy J,
Kristinsson S,
Healy A,
Dalbow G,
Goodenough D,
Pawlicki T
Publication year - 2013
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 - 10.1118/1.4813976
Subject(s) - scanner , image quality , statistical process control , computer science , standard deviation , image processing , six sigma , reliability engineering , software , process (computing) , statistics , data mining , artificial intelligence , mathematics , image (mathematics) , engineering , chemical engineering , cascade , programming language , operating system
Purpose: Statistical Process Control is regularly used in industry to refine operations and improve quality. Historically the measurement of CT scanners has only been based on pass/fail criteria with minimal understanding of data fluctuations and implications of these fluctuations as possible early warning systems. We review different data sets in an effort to better understand how these tools can be used to establish tighter performance controls to better indicate potential system failures. Methods: We investigate effective energy, slice thickness, uniformity and resolution from data collected from weekly QA tests on one CT scanner (Site 2) over the course of one year. For comparison we use a previously collected set of 1565 scans from a ten year period during a quantitative imaging study (Site 1). The image data was processed with the ImageOwl CatphanQA software. Tolerances for CT imaging quality parameters were selected from IEC‐61223‐3‐5 and other sources. Process capability, Cp, describes the ratio of tolerance range to observed standard deviation. Cpk additionally weighs in closeness to one side of specification limits. For a Gaussian variable, process capability equal to 1 means 99.7% chance of correctly observing a single measurement within specification, i.e. 3 sigma. Results: Process capabilities are better than six sigma (Cp>2) for slice thickness measurements and sensitometry for Site 1, but not for uniformity. The greatest difference between sites were observed in sensitometric readings. The data from Site 1 showed a higher variability in slice thickness during the first year of operation, where it was discovered that the Field Of View (FOV) was often too large. This FOV check has been implemented as an automatic filter tool in the ImageOwl CatphanQA software. Conclusion: By monitoring modern CT scanners with today's test systems using commonly accepted control limits, reliable go/no‐go signals can be generated for most CT QA variables. ImageOwl sells the cloud‐based image QA software used in the study and pays the salary for some of the authors