ESTRO 2023 - Abstract Book

S129

Saturday 13 May

ESTRO 2023

distributions from cancerous and non-tumoral regions of diagnostic pre-treatment hematoxylin and eosin-stained digital histopathology whole slide images were extracted. Each gigapixel histopathology whole slide (WSI) image was separated into patches. For each patient's WSI, the nuclei center was automatically detected by a blob detection algorithm where each nuclear content area was computed generating a nuclei radius distribution, a Voronoi cell diagram was constructed to estimate cell distributions. Probability of post radiotherapy recurrence given a pre-treatment patient-specific nuclei distribution mean was computed, as well as student t-tests for clinical variables such as stage, cancer type, and age at diagnosis between patients who had opposite post radiotherapy recurrence outcomes. Results Across all patients, inter-patient variability in nuclei distribution mean was observed for the both non-tumoral and cancerous regions with a median of 1.729 microns and 2.837 microns respectively, as presented in Figure 1. On a patient- level, a decrease in the probability of post radiotherapy recurrence given a larger patient-specific nuclei distribution mean was observed for both cancerous and non-tumoral region features as shown in Figure 2. Similarly in radiobiology experiments, Behmand et.al (Behmand et al 2021) showed a strong positive correlation between nuclei of HeLa and PC3 cell line with the number of γ -H2AX foci at an absorbed dose of 1, 2, and 4 Gy following irradiation with 225 kV x-rays and iridium 192 high dose rate brachytherapy. This offers a quantitative radiobiological explanation for the decreased likelihood in post radiotherapy probability for patients with larger nuclei.

Conclusion Larger patient specific nuclei may cause more DNA damage on a radiobiological standpoint, which translated to lowered probability of post radiotherapy recurrence on a patient level. The results illustrate the importance of multiscale dosimetry where variation in patient specific nuclei were taken into account to enable personalized dose prescriptions and achieve optimal treatment prognosis in the clinic. PD-0169 An innovative and versatile deep learning approach to estimate the out-of-field dose. N. Benzazon 1 , A. Carré 1 , F. de Kermenguy 1 , R. Allodji 2 , F. de Vathaire 2 , E. Deutsch 1 , I. Diallo 1 , C. Robert 1 1 Gustave Roussy, U1030, Villejuif, France; 2 Gustave Roussy, U1018, Villejuif, France Purpose or Objective Radiation therapy has common iatrogenic effects, including the development of radiation-induced lymphopenia, radiation- induced cancer, and cardiac and vascular complications. A growing body of scientific evidence reveals the potential effects of medium (Gy) and low doses (< 1 Gy) in particular on highly radiosensitive immune cells. Therefore, the assessment of low doses inevitably delivered outside the treatment field (out-of-field dose) is a topic of renewed interest at a pivotal moment in the development of combination therapies. In this work, we propose an original fast and versatile tool to estimate out-of-field doses for patients treated with external photon radiotherapy with energies above 1 MV based on a Deep Learning approach.