ESTRO 2025 - Abstract Book

S3799

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Results: Nine mng experiencing a local relapse (relapse mng) after a median time of 37 months were evaluated along with eight patients gaining local control after IMPT, overall accounting for 47 follow-up exams. The median validation auto contouring dice-score was 0.64 hence, due to sub-optimal segmentation performance, 12 follow-up segmentations (26%) needed manual adjustments. Besides the significant increase of the whole lesion volume at relapse follow-up (p=0.002), low-contrast and intermediate contrast regions within the GTV reported a significantly lower increase (+11% vs. +35%, p=0.011) and increase (+9% vs. -5%, p=0.031), respectively, compared to non-relapse follow-ups. Moreover, non-relapse follow-up of relapse mng reported significantly increasing intermediate contrast region volumes (+2% vs. -29%, p=0.002) within GTV, although with lower enhanced-T1 values (-6% vs. +6%, p=0.009) compared to non-relapse mng. Conclusion: Our preliminary results showed accordance between the described technical framework and the mng follow-up radiological evaluations, hence representing a potential tool to aid radiologists improving their efficiency and standardizing quantitative reporting. Nonetheless, auto-contouring models with higher performance should be further explored, as well as comprehensive models for early detection of IMPT relapses developed and tested on larger dataset.

Keywords: MRI, quantitative, clustering

References: [1]

Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep learning based biomedical image segmentation. Nat Methods 2021;18:203–11. https://doi.org/10.1038/s41592-020-01008-z.