Segmentation on low resolution magnetic resonance images for prostate radiotherapy
PO-1691
Abstract
Segmentation on low resolution magnetic resonance images for prostate radiotherapy
Authors: Samuel Fransson1, Robin Strand2, David Tilly1
1Uppsala University Hospital, Medical Physics, Uppsala, Sweden; 2Uppsala University, Information Technology, Uppsala, Sweden
Show Affiliations
Hide Affiliations
Purpose or Objective
Real-time tracking on an MR-Linac requires 3D-segmentation with high temporal resolution as well as high spatial resolution. MR-imaging generally prohibits this in 3D and currently only 2D-acquisition is possible at such high temporal resolution and thereby the full information about the motion cannot be determined. In this work, we aim to solve this problem with high-temporal resolution 3D-image acquisition by sacrificing the spatial resolution of the images. We then trained neural networks to obtain the high-resolution segmentations with these low-resolution images as input, thus investigating the performance of such segmentation approaches based on different levels of spatial resolution.
Material and Methods
A total of 163 T2-weighted MR-scans of 26 prostate patients treated at our MR-Linac were included along with manual segmentations of CTV, bladder and rectum. The images were downsampled into three levels, reaching 1/2, 1/4 and 1/8 of the initial resolution in the phase-encoding directions while maintaining the resolution in the frequency encoding direction, thus mimicking a reduction in acquisition time of 4, 16, and 64, respectively. For each level as well as the original high-resolution data, a 2D U-net deep learning framework was trained on 20 patients scans (total 128 3D-images) to segment the high resolution structures, and evaluated on the remaining 6 patients scans (total 35 3D-images) and comparing with the ground truth manual segmentations.
Results
The DICE-coefficient and the 95% Hausdorff distance for the evaluation dataset is depicted in Table 1 as the mean and the standard deviation.
Table 1.
| Resolution level \ Structure | CTV DICE Mean(std) HD 95% [mm] Mean(std) | Bladder DICE Mean(std) HD 95% [mm] Mean(std) | Rectum DICE Mean(std) HD 95% [mm] Mean(std) |
| 1/1 | 0.85(2) 5(1) | 0.93(12) 3.9(4.8) | 0.84(7) 15.8(15.9) |
| 1/2 | 0.86(3) 4.8(1.3) | 0.93(3) 4.7(3.5) | 0.84(6) 16.3(17.2) |
| 1/4 | 0.84(3) 5.5(1.6) | 0.93(6) 3.9(2.7) | 0.84(6) 15.1(16.8) |
| 1/8 | 0.80(5) 7.9(3.2) | 0.92(5) 4.8(2.8) | 0.82(7) 15.6(15.7) |
Conclusion
The rather small differences in both DICE and hausdorff distance indicates that the resolution level has small impact on the overall performance. A small difference can be seen for the lowest resolution and the CTV though, which may be considered the most important structure since it is the target. Although only small differences between resolutions, the hausdorff distances for the rectum stands out with high values, likely due to the utilization of 2D-networks struggling to determine the extent in the craniocaudal direction, and possibly also due to inconsistent ground truth segmentation. Although a study based on simulated data, the results indicate a possibility for spatial resolution reduction while maintaining segmentation performance.