Encrypted login | home

Program Information

Intra-Patient Deformable Image Registration Uncertainties Quantified Using the Distance Discordance Metric


Z Saleh

Z Saleh1*, M Thor1 , A Apte1 , G Sharp2 , L Muren3 , J Deasy1 , (1) Memorial Sloan Kettering Cancer Center, New York City, New York, (2) Massachusetts General Hospital, Boston, MA, (3) Aarhus University Hospital, Aarhus, Denmark,

Presentations

SU-E-J-159 Sunday 3:00PM - 6:00PM Room: Exhibit Hall

Purpose:
The quantitative evaluation of deformable image registration (DIR) is currently challenging due to lack of a ground truth. In this study we test a new method proposed for quantifying multiple-image based DIR-related uncertainties, for DIR of pelvic images.

Methods:
19 patients were analyzed, each with 6 CT scans, who previously had radiotherapy for prostate cancer. Manually delineated structures for rectum and bladder, which served as ground truth structures, were delineated on the planning CT and each subsequent scan. For each patient, voxel-by-voxel DIR-related uncertainties were evaluated, following B-spline based DIR, by applying a previously developed metric, the distance discordance metric (DDM; Saleh et al., PMB (2014) 59:733). The DDM map was superimposed on the first acquired CT scan and DDM statistics were assessed, also relative to two metrics estimating the agreement between the propagated and the manually delineated structures.

Results:
The highest DDM values which correspond to greatest spatial uncertainties were observed near the body surface and in the bowel due to the presence of gas. The mean rectal and bladder DDM values ranged from 1.1-11.1 mm and 1.5-12.7 mm, respectively. There was a strong correlation in the DDMs between the rectum and bladder (Pearson R = 0.68 for the max DDM). For both structures, DDM was correlated with the ratio between the DIR-propagated and manually delineated volumes (R = 0.74 for the max rectal DDM). The maximum rectal DDM was negatively correlated with the Dice Similarity Coefficient between the propagated and the manually delineated volumes (R= -0.52).

Conclusion:
The multiple-image based DDM map quantified considerable DIR variability across different structures and among patients. Besides using the DDM for quantifying DIR-related uncertainties it could potentially be used to adjust for uncertainties in DIR-based accumulated dose distributions.


Contact Email: