ESTRO 2023 - Abstract Book

S1879

Digital Posters

ESTRO 2023

1 Univ Rennes, Inserm, LTSI - UMR 1099, Rennes, France; 2 CLCC Eugène Marquis, Radiation Oncology, Rennes, France

Purpose or Objective Prostate cancer has been typically treated with a total radiation dose of 74-80 Gy administered in 2 Gy fractions during 7 or 8 weeks from Monday to Friday. In the recent years, moderate hypofractionated schedules have been proposed and tested in randomised clinical trials. Results suggest that hypofractionated radiotherapy may be recommended as a new standard for localised prostate cancer. Nevertheless, numerous uncertainties about the dose equivalences, considering both local tumour control and toxicity, still exist. The objective of this work was thus to use a previously developed and validated in silico model of tumour response to identify dose equivalences between 2, 2.5, and 3 Gy fractionations in terms of freedom from biochemical recurrence. Materials and Methods A cohort of 279 patients from the CLCC Eugène Marquis with localised prosate cancer and treated with EBRT was used for this study. A pre-treatment 3 T MRI scan was obtained for every individual. Sequences included T2-w. A standard hypofractionation schedule (2 Gy per fraction, with a total dose of 74-80 Gy) was administered to every patient. In order to simulate the tumour response to radiotherapy with our in silico model, 279 2D virtual tissues representing the 279 patients of the cohort were built. Tumour volumes and average T2-w intensities were extracted from the corresponding pre-treatment MRI scans and then used to initialise the virtual tissues. Then, the standard irradiation schedule administered to each patient of the cohort was simulated on the corresponding digital tissue. The number of virtual tumour cells at the end of treatment (8 weeks) was obtained. In previous work, it has already been shown that this simulation output is highly predictive of biochemical recurrence. Finally, alternative 2.5 and 3 Gy fractionations (also 5 sessions per week from Monday to Friday) were explored for every patient. Virtual tissues were irradiated until they contained the same number of virtual tumour cells that after the 2 Gy schedule. Total doses needed to obtain these numbers of tumour cells were identified. The simulation was repeated 5 times on each virtual tissue and the mean output value was taken. Results Equivalent total doses between the three irradiation schedules are presented in Fig. 1. Median total doses of 70.6 and 64.5 Gy, were obtained for the 2.5 and 3 Gy fractionations, respectively. These results, obtained through in silico simulation, are in line with dose equivalences reported in clinical trials.

Fig 1. Equivalent total doses. Total doses for the 2 Gy fractionation correspond to those administered to the cohort. Total doses for the 2.5 and 3 Gy fractionations were obtained through in silico simulation. Conclusion Total dose equivalences identified by our mechanistic model of tumour response were congruent with the results reported in the clinical literature. This work paves thus the way for the identification in silico of dose equivalences between hypofractionated schedules not explored yet in clinical trials.

PO-2096 Machine learning prediction of Dice similarity coefficient for accuracy evaluation

Y.M. Wong 1 , P.L. Yeap 2 , A.L.K. Ong 2 , H.Q. Tan 2 , W.S. Lew 1 , J.C.L. Lee 1,2

1 Nanyang Technological University, School of Physical and Mathematical Sciences, Singapore, Singapore; 2 National Cancer Centre Singapore, Division of Radiation Oncology, Singapore, Singapore Purpose or Objective Following the advent of highly conformal radiotherapy techniques, patient anatomic variations are having a greater impact on the daily dose distributions. This calls for regular adjustment of treatment plan, a process known as adaptive radiotherapy (ART). Deformable image registration (DIR), a technique to transform one image to another, is indispensable in an ART workflow. While contour-based metrics, e.g. Dice similarity coefficient (DSC), are commonly used for DIR accuracy evaluation, they require the manual delineation of contours, which is a bottleneck in the radiotherapy workflow due to its laborious nature. In this work, we presented a novel method of predicting DSC using DVF-based metrics by applying several machine learning models, to achieve a quick DIR validation process without much human intervention.