Purpose: To design a robust and fully automated image processing algorithm for accurately analyzing star shot (also known as ‘spoke shot’) images, as recommended by the AAPM TG-142 guidelines. The proposed algorithm was specifically designed to eliminate external inputs (i.e., 100% automated) with a focus on generalizability for maintaining increased accuracy across a multitude of highly variable beam characteristics and image qualities, while also providing a significant time-save in the clinic.
Methods: A fully automated ad hoc algorithm was designed for accurately and reliably analyzing star shot images without the need for external inputs. Performance of the algorithm was tested across a set of 29 star shot images. The images were comprised of a total of 165 beams. Analysis was performed on both a per-image and a per-beam basis. Generalizability was demonstrated with an array of gantry, collimator, and couch star shot configurations typically found in the clinical setting. In addition, because beam properties are likely to vary within the clinical settings, whether due to machine restraints, unique procedures, or preferences of the physicist themselves, the algorithm was designed to be invariant to beam widths, number of beams, and relative beam doses.
Results: The proposed algorithm correctly analyzed all beams in 28 of the 29 star shot images (96.6%). Additionally, of the 165 beams, 163 (98.8%) were correctly identified. The single image that failed served as a test case for the resolution of the image processing algorithm as it could not resolve two beams with a large proportion of overlapping dose.
Conclusion: This work demonstrated a robust solution for fully automating the star shot image analysis process with high accuracy and reliability, while also saving valuable time. Such automation provides a vastly improved workflow for medical physicists to perform necessary QA tasks more efficiently.
Not Applicable / None Entered.