(Wednesday, 8/1/2018) 4:30 PM - 6:00 PM
Room: Karl Dean Ballroom A1
Purpose: The use of algorithms to perform autonomous image QA analysis using quality assurance (QA) phantoms has many advantages over human based manual inspection. These include more quantitative data collection, reduced user bias, and less human effort. Unfortunately, such QA algorithms are phantom specific, in that they are hardcoded at the programming level to only identify features of well-known geometries. This therefore greatly limits the end user in the choice of phantoms to use in their clinic. To overcome these limitations, we present an algorithm that is aimed to be universally compatible with any pre-existing QA phantom.
Methods: A XML-based vector image format language was modified for inclusion of several new tags that describe functional features (contrast, resoultion, etc.) in addition to spatial features (size, shape, location, etc.) of QA objects contained in a phantom. A python based program (UniPhan), was developed to decode these a XML wireframes, automatically register them to a DICOM image, perform image analysis, and output QA results. Current functional objects supported by Uniphan include contrasts regions, ROI intensity/noise, light-radiation field, MTF bars/edges/points, linearity, and uniformity.
Results: Uniphan was clinically used over a 3 year period using ACR464, Catphan, LasVegas, Isoalign, and TOR18FG phantoms for performing various TG-142 QA tasks. A total of 590 monthly QA tasks were performed involving a number of end users. QA results produced were found to be consistent with results from several commercial QA software packages.
Conclusion: A novel imaging QA algorithm was developed that is not phantom specific. XML based wireframes can be created using common SVG graphics editing software allowing wide spread use. The algorithm performed well for a variety of different phantoms and is currently used in a clinical environment.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
