Program Information
Semi-Automated in Vivo Radiochromic Film Dosimetry Using a Novel Image Processing Algorithm
M Reyhan*, N Yue , Rutgers University, New Brunswick, NJ
Presentations
SU-E-T-497 Sunday 3:00PM - 6:00PM Room: Exhibit HallPurpose:To validate an automated image processing algorithm designed to detect the center of radiochromic film used for in vivo film dosimetry against the current gold standard of manual selection.
Methods:An image processing algorithm was developed to automatically select the region of interest (ROI) in *.tiff images that contain multiple pieces of radiochromic film (0.5x1.3cm²). After a user has linked a calibration file to the processing algorithm and selected a *.tiff file for processing, an ROI is automatically detected for all films by a combination of thresholding and erosion, which removes edges and any additional markings for orientation. Calibration is applied to the mean pixel values from the ROIs and a *.tiff image is output displaying the original image with an overlay of the ROIs and the measured doses. Validation of the algorithm was determined by comparing in vivo dose determined using the current gold standard (manually drawn ROIs) versus automated ROIs for n=420 scanned films. Bland-Altman analysis, paired t-test, and linear regression were performed to demonstrate agreement between the processes.
Results:The measured doses ranged from 0.2-886.6cGy. Bland-Altman analysis of the two techniques (automatic minus manual) revealed a bias of -0.28cGy and a 95% confidence interval of (5.5cGy,-6.1cGy). These values demonstrate excellent agreement between the two techniques. Paired t-test results showed no statistical differences between the two techniques, p=0.98. Linear regression with a forced zero intercept demonstrated that Automatic=0.997*Manual, with a Pearson correlation coefficient of 0.999. The minimal differences between the two techniques may be explained by the fact that the hand drawn ROIs were not identical to the automatically selected ones. The average processing time was 6.7seconds in Matlab on an IntelCore2Duo processor.
Conclusion:An automated image processing algorithm has been developed and validated, which will help minimize user interaction and processing time of radiochromic film used for in vivo dosimetry.
Contact Email: