ESTRO 2022 - Abstract Book

S768

Abstract book

ESTRO 2022

Conclusion The machine learning system was able to find relevant deviations based only on local delineation practice. More than half of the outliers were classified as true outliers. Based on these results, clinical implementation at the MR linac has been initiated.

MO-0880 Quantifying individual dose relationships for pCR in rectal cancer: potentials for a customized dose

A. Cicchetti 1 , P. Passoni 2 , S. Broggi 1 , N. Slim 2 , A. Del Vecchio 1 , S. Cascinu 3 , F. De Cobelli 4 , R. Rosati 5 , N. Di Muzio 2 , C. Fiorino 1 1 San Raffaele Scientific Institute, Medical Physics, Milan, Italy; 2 San Raffaele Scientific Institute, Radiotherapy, Milan, Italy; 3 San Raffaele Scientific Institute, Oncology, Milan, Italy; 4 San Raffaele Scientific Institute, Radiology, Milan, Italy; 5 San Raffaele Scientific Institute, Gastroenterology Surgery, Milan, Italy Purpose or Objective To develop a model for individual pathological Complete Response (pCR) prediction for rectal cancer patients (pts) treated with neo-adjuvant radio-chemotherapy (RCT). The suggested approach combines the classical Logit dose-effect curve with the individual early response assessed by MRI imaging during RCT. Materials and Methods A population-averaged dose-effect curve for pCR may be derived by fitting published data: the recent meta-analysis by Hall et al. was utilized. Then, we considered data of 90 pts from our Institute treated with RCT in which a relationship between a previously introduced early regression index (ERI, assessed at half RCT) and pCR was quantified and found to be highly predictive. The index is proportional to the ratio of tumor volumes during and before RCT, with lower values indicating better response. ERI was considered as a dose modifying factor (DMF), "adapting" the resulting dose-effect curve to this population. The Logit curve assessed from the meta-analysis was first tuned on our population: the bivariate logistic pCR model was converted into a Logit EQD2 model with ERI acting as a dose modifying factor (DMF) for the estimated rate of pCR. The final model resulted in a series of sigmoid shaped curves, depending on individual ERI values. After successful fitting, the model was applied to quantify the expected increase of pCR with dose, depending on patient response. Results The resulting Logit model depending on the 2Gy-equivalent dose (EQD2) and ERI showed best-fit values of TD50 = 52.2 Gy, DMF = 0.017 x ERI + 0.89 the steepness m = 1/(3.7 + 0.07 x ERI). Of note, the meaning of TD50 coincides with the classical one (i.e.: the dose corresponding to pCR=50%) for ERI=6.9, corresponding to a quite highly responding pts in terms of tumor regression.