ESTRO 2024 - Abstract Book

S3095

Physics - Autosegmentation

ESTRO 2024

Figure 1. Trimap Diagram

Results:

Our evaluation consisted of both qualitative and quantitative analyses. In qualitative analysis, we visually appraised the algorithm's capacity to enhance tumor boundaries by comparing results with input images to gauge its objectivity. In the qualitative assessment, the algorithm consistently produced tumor boundaries that aligned with the input images. Notably, the algorithm effectively utilized offline PET data to establish precise boundaries for ROIs that presented challenges due to less distinct boundaries in daily MRIs. In addition, we conducted a quantitative assessment by calculating Dice scores through confidence image binarization, using manual delineation as the ground truth. Quantitatively, the algorithm achieved an average Dice score of 0.89 for GTV and 0.77 for DIL, outperforming DL-based methods, which achieved 0.82 and 0.75, respectively.

Conclusion:

Our approach effectively enhances DIL and GTV boundary. Of particular importance is the algorithm's objectivity, especially in segments of the ROI that are challenging to discern. This objectivity holds great promise for augmenting the accuracy and reliability of SIB treatment planning in prostate SBRT in the context of ART.

Keywords: Prostate, Dominant Intraprostatic Lesion, MRI

2191

Digital Poster

Automatic segmentation of the prostate, urethra, and prostatic zones

William Holmlund 1 , Attila Simkó 1 , Karin Söderkvist 1 , Péter Palásti 2 , Szilvia Tótin 2 , Kamilla Kalmár 2 , Zsófia Domoki 2 , Zsuzsanna Fejes 2 , Tamás Z Kincses 2 , Patrik Brynolfsson 1,3 , Tufve Nyholm 1