ESTRO 2025 - Abstract Book

S3433

Physics - Machine learning models and clinical applications

ESTRO 2025

Conclusion: We developed VerGAReg, a digital anatomy-based vertebra-guided affine registration network that demonstrates superior performance in abdominal imaging. VerGAReg can serve as an effective preliminary step for automatic affine registration in abdominal applications.

Keywords: cross-modality, dual-channel network

References: [1] L. Xing et al. , "Overview of image-guided radiation therapy," Medical Dosimetry, vol. 31, no. 2, pp. 91-112, 2006. [2] M. Y. Evan, A. Q. Wang, A. V. Dalca, and M. R. Sabuncu, "KeyMorph: Robust multi-modal affine registration via unsupervised keypoint detection," in International Conference on Medical Imaging with Deep Learning , 2022, pp. 1482-1503: PMLR. [3] W. P. Segars, M. Mahesh, T. J. Beck, E. C. Frey, and B. M. Tsui, "Realistic CT simulation using the 4D XCAT phantom," Medical physics, vol. 35, no. 8, pp. 3800-3808, 2008.

3716

Poster Discussion Validation of a combined ERI-Dose Model for Predicting Pathological Complete Response in Rectal Cancer Alessandro Cicchetti 1 , Giuditta Chiloro 2 , Davide Cusuamano 3 , Sara Broggi 4 , Martina Mori 4 , Paolo Passoni 5 , Nadia Gisella Di Muzio 5 , Lorenzo Placidi 2 , Angela Romano 2 , Matteo Nardini 2 , Luca Boldrini 2 , Maria Grazia Gambacorta 2 , Claudio Fiorino 4 1 Data Science Unit, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. 2 Dep. of Human Oncology, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy. 3 Medical Physics Dep, Mater Olbia Hospital, Olbia, Italy. 4 Medical Physics Dep, IRCCS San Raffaele Scientific Institute, Milan, Italy. 5 Radiotherapy Unit, IRCCS San Raffaele Scientific Institute, Milan, Italy Purpose/Objective: To validate a model predicting pathological Complete Response (pCR) in rectal cancer patients [1] based on a combined metric of Early Regression Index (ERI) and prescribed tumour dose, using data from two distinct RT protocols.

Material/Methods: We want to validate the following function computed using dose data from [2] and ERI dependencies from [3]:

with TD50=52Gy ( EQD2 with α/β=10Gy and γ=0.6Gy); Dose Modifying Factor (DMF) = (0.0166* ERI +0.8858); Steepness(m)= 1/(3.7+0.07* ERI ).

The cohort included 132 patients from an external Institute. Of these, 68 patients were treated with SIB delivering 55Gy/25fr to the GTV (EQD2=57.4Gy). ERI was retrospectively calculated at mid-RT using MRI on an MR-Linac for these patients. The remaining 64 patients were part of a clinical trial involving adaptive RT, where patients with ERI>13.1 received a dose escalation (boost regimen: 10 fractions at 2.2Gy followed by 15 fractions at 2.57Gy, EQD2=65.9Gy). Patients with ERI≤13.1 continued the standard SIB treatment. The median ERI for the cohort was 15 (IQR=9.1-31.8). Calibration plots, AUC, and Average Precision-AUC (PR-AUC) metrics were used to assess model performance. Recalibration was also performed to refine predictions.