Abstract
Purpose:
Prostate MRI plays an important role in diagnosis, biopsy guidance, and therapy planning for prostate cancer. Prostate MRI contours can be used to aid in image fusion for ultrasound biopsy guidance and delivery of radiation. Our goal in this study is to evaluate an automatic atlas‐based segmentation method for generating prostate and peripheral zone (PZ) contours on MRI.
Methods:
T2‐weighted MRIs were acquired on 3T‐Discovery MR750 System (GE, Milwaukee). The Volumes of Interest (VOIs): prostate and PZ were outlined by an expert radiation oncologist and used to create an atlas library for atlas‐based segmentation. The atlas‐segmentation accuracy was evaluated using a leave‐one‐out analysis. The method involved automatically finding the atlas subject that best matched the test subject followed by a normalized intensity‐based free‐form deformable registration of the atlas subject to the test subject. The prostate and PZ contours were transformed to the test subject using the same deformation. For each test subject the three best matches were used and the final contour was combined using Majority Vote. The atlas‐segmentation process was fully automatic. Dice similarity coefficients (DSC) and mean Hausdorff values were used for comparison.
Results:
VOIs contours were available for 28 subjects. For the prostate, the atlas‐based segmentation method resulted in an average DSC of 0.88+/−0.08 and a mean Hausdorff distance of 1.1+/−0.9mm. The number of patients (#) in DSC ranges are as follows: 0.60–0.69(1), 0.70–0.79(2), 0.80–0.89(13), >0.89(11). For the PZ, the average DSC was 0.72+/−0.17 and average Hausdorff of 0.9+/−0.9mm. The number of patients (#) in DSC ranges are as follows: <0.60(4), 0.60–0.69(6), 0.70–0.79(7), 0.80–0.89(9), >0.89(1).
Conclusion:
The MRI atlas‐based segmentation method achieved good results for both the whole prostate and PZ compared to expert defined VOIs. The technique is fast, fully automatic, and has the potential to provide significant time savings for prostate VOI definition.
AS Nelson and J Piper are partial owners of MIM Software, Inc. AS Nelson, J Piper, K Curry, and A Swallen are current employees at MIM Software, Inc.
