ESTRO 2020 Abstract book

S1069 ESTRO 2020

overestimation in loss of clonogenicity with dose by up to a factor of 4. To conclude, we have found substantial evidence that time-lapse analysis, enhanced by machine learning algorithms and complemented with a biomathematical model reveals a greater breadth of cellular responses to radiation than fixed-time assays. PO‐1824 Superparamagnetic iron oxide nanoparticles as radiosensitisers with magnetic fields. E. Russell 1 , M. Barry 2 , S. Osman 1 , G. Schettino 2 , C. McGarry 3 , K. Prise 1 1 Queen's University Belfast, Centre for Cancer Research and Cell Biology, Belfast, United Kingdom ; 2 National Physical Laboratory, Radiation Biology, London, United Kingdom ; 3 Belfast Health and Social Care Trust, Radiotherapy Physics, Belfast, United Kingdom Purpose or Objective MR-Linacs allow radiotherapy treatments to be carried out alongside MRI imaging for improved soft tissues image guidance. Contrast agents are often required in MRI to improve tumour visibility. Iron oxide nanoparticles (SPIONs) are being investigated as radiosensitisers for their preferential uptake in tumours, and their ability to absorb more radiation, causing more DNA damage to the tumour. The aim of this investigation is to assess the visibility of SPIONs in T1-weighted MRI and the effect of SPIONs with (a) 225kVp and (b) 6MV x-rays with and without a static Four cancer cell lines were selected; H460 (lung), DU145 (prostate), MiaPaCa2 (pancreas), and U87 (brain). Clonogenic assays were performed using a 225 kVp x-ray source in combination with 0.1 mM of 5 nm SPIONs (Fe 3 O 4 ) with dose levels of 4 Gy and 8 Gy. Cells were exposed to the SPIONs for 24 hrs prior to radiation. After an incubation period of 7 days, the number of colonies formed was counted, and the Dose Enhancement Factor (DEF) calculated using the plotted clonogenic cell survival curves. H460 cells were taken forward to receive 6 MV radiation in combination with a 1.5 T static magnetic field. Immunofluorescence assays were used to determine the level of double-strand breaks in the DNA caused by radiation and SPIONs by labelling 53BP1 antibodies 1hr and 24 hr post irradiation. Results Radiosensitisation from SPIONs appears to be cell line specific, with H460 cells showing a Dose Enhancement Factor (DEF) of 1.33 at 8Gy, using 225kVp x-rays. There was no significant change in surviving fraction for all other cell lines. In combination with a 1.5T magnetic field, at 6MV, a significant increase in DNA damage was seen with H460 cells in nanoparticles, both 1hr and 24hr post irradiation, with an average number of foci found as 25.2±2.46 and 5.17±1.70 respectively, compared to controls, shown in Figure 1. However, a small but significant change between 0T and 1.5T was only seen at 24hr (P<0.0001). Imaging signal form the SPIONs increased linearly with increasing concentration, shown in Figure 2. 1.5T magnetic field. Material and Methods

intervals for 9 days. We developed an automated image segmentation routine that identifies individual colonies and records morphological parameters. We tracked colonies over time by pairwise matching on subsequent images. We then identified exponentially growing colonies and extracted CG parameters by using wavelet decomposition to identify periods of exponential CG and by training a tree ensemble classifier on a combination of morphological, wavelet and curve-fitting parameters. Based on the extracted CG data, we developed a biomathematical stochastic model that explicitly calculates the effects of radiation damage, colony growth, -stagnation, and -fusion to predict colony size distributions as a function of dose and time. Results The image segmentation algorithm achieves an average accuracy of 99 % (97-99 %). Colony classification displays an average accuracy of 85 % (78-91 %). Between 57 % and 95 % of the colonies exhibit phases of exponential CG, depending on radiation dose and cell line. CG rates are normally distributed with large standard deviations of up to 80 % and decrease with increasing dose. We observed a large variability of these parameters between the cell lines (see Fig. 1).

Comparing predictions from our model to the experimental data confirmed the hypothesis that stochastic CG explains the observations (see Fig. 2). Note that our model uses no adjustable parameters but only the extracted parameters on CG rate distributions.

Conclusion The observed decrease in CG rate with dose presents a potentially serious source of systematic error in clonogenicity scoring: in our study, it leads to an