ESTRO 2024 - Abstract Book

S3955

Physics - Image acquisition and processing

ESTRO 2024

Five patients received planning MR images in the radiotherapy position. Standard and DL reconstructed T2-weighted MR images were acquired. The DL images were optimised to improve image quality and reduce acquisition time. Two experienced delineators independently contoured the prostate, seminal vesicles, bladder, rectum and penile bulb on both images, blinded to image type. They recorded the time to contour each organ and rated the visibility of each organ boundary on a 4-point scale: 1-Impossible to discern, 2-Hard to discern, 3-Moderately discernible, 4-Clearly discernible. Inter-delineator variability was quantified using mean distance to agreement (DTA). Following this, automatic contours were also created on both images using a DL T2 male pelvis model (MVision). The time required to generate contours on one image was recorded. Each delineator reviewed and edited each contour, timing how long that took and also rated the contour on a 5-point scale: 1-Complete recontouring, 2-Significant correction, 3-Some correction, 4-Minor corrections (acceptable), 5-Acceptable.

Results:

The overall visual appearance of the DL images was improved (figure 1) and acquisition time reduced from 4.3 to 3.4 minutes. Delineator ratings were better for all organs on the DL images, although the differences were small except for the prostate and seminal vesicles (table 1). One DL image had significant motion artefact which produced a clear outlier in the ratings (figure 2). The improved ratings did not translate into reduced manual delineation timing except for the seminal vesicles which reduced by 1.7 minutes. There was no change in the inter-delineator variability except for the penile bulb which was unexpectedly 1.5 mm worse for the DL image.