ESTRO 2024 - Abstract Book

S5142

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

For all reference anatomies and analysis, the largest region was identified for the physical sub-scale, covering parts of the brain (including the pituitary gland, brainstem and cerebellum), part of the pharyngeal constrictor muscles (PCMs; at the level of the C3 vertebra) and the larynx (at the level of the thyroid cartilage). The second largest region was for the composite MDADI score, covering the lowest part of the brainstem (medulla oblongata), PCMs and larynx (the latter two in similar locations as identified with the physical subscale). Next, the region identified for the functional subscale comprised the PCMs and larynx. Last, very small regions were identified for the emotional subscale. Comparing the conventional versus the bootstrapped IBDM, we observed that bootstrapping increased all regions in volume, and the shape of some regions (e.g., physical subscale) is slightly more consistent between reference patients, indicating a potential benefit of bootstrapping for region identification in small cohorts. We investigated the impact of MDADI subscales in spatial outcome modelling using both conventional and bootstrapped IBDM. The observed differences between regions indicates that independent analysis of the subscales can provide complementary and differential information [5] . In particular, the physical subscale should be further studied in a larger cohort to explore/confirm the links between late dysphagia post-radiotherapy and dose to the identified anatomical regions. The roles of the pharyngeal constrictor muscles, larynx and brainstem have been already identified [4,6,7] , however the role for the cerebellum, a substructure known to be involved in swallowing [8] , has not been previously reported. Moreover, we observed little to no correlation for the emotional subscale, re-enforcing the recommendation to drive spatial dose outcome modelling using objective physiological/functional measures [4,9] . Bootstrapped IBDM showed potential to improve consistency warranting further research varying the number of resamples to reach robust identification of significant regions for small cohorts. Conclusion: