ESTRO 38 Abstract book

S1029 ESTRO 38

A heterogeneous tumour is assumed to consist of the following mixture of cells varying only in α : α i = [0.1;0.15;0.2;0.25;0.3;0.35] Gy -1 , log(N o,i ) = [3.7; 4; 4.3; 5; 7; 8] with β = 0.02 Gy -1 ; sub-population ‘i’ consists of initial cell number N o,i with radiosensitivity α i . For this tumour a theoretical cell survival curve is constructed according to the linear-quadratic (LQ) model of cell killing. Also а TCP vs (total) dose curve is constructed based on the Poisson TCP model for n = 20 equal fractions. Both curves consist of 8 points corresponding to different doses. The error in the cell-survival data was considered to be 10% and the error in the TCP data 3%. In this study the two data-sets are then simultaneously fitted with a single-component (homogeneous) LQ model of cell killing and a single-component Poisson TCP model respectively. The χ 2 method of fitting is used on whose basis a p-value is calculated serving as a measure of the goodness of the fit. Results Two different versions of fitting were used. It was firstly assumed that the two data sets may be described with one unique value of the radiosensitivity α. The result is shown in figs 1a) and 1c) respectively. The p-value is 0. In figs 1b) and 1d) are shown a fit of the LQ model to the pseudo- experimental cell-survival curve and a simultaneous fit of the TCP model to the TCP curve, where two separate values of α for the two models are assumed. The p-value of the fit is 1. The pseudo-experimental data points are shown in the figures with their assumed error bars.

Purpose or Objective Cell apoptosis plays critical roles in cancer research and treatment. The most commonly used methods for high- throughput detection of cell apoptosis are based on flow cytometry (FCM) in combination with the dyes which bond to different components in the cells. Despite of the accuracy, these methods are characterized with higher cost and time consuming associated with staining cells. Here we represent a stain-free method based on diffraction imaging flow cytometry and machine learning techniques for detecting cell in different apoptosis stages. Material and Methods For apoptosis induction, human erythroleukemia K562 cells were treated with cisplatin (DDP) in a concentration of 25 mg/ml for 24. And the cell samples without any treatment were used as the control group. Annexin V-PE and SYTOX® Green Dead cell stain were used for cell sorting using fluorescence activated cell sorting (FACS). After cell sorting K562 cells were separated into three sub- groups which were viable, early apoptotic, and late apoptotic/necrotic cells. These sub-groups were measured with our polarization diffraction imaging flow cytometry (p-DIFC) system and the cell images were acquired. In processing the image pairs with s and p polarizations of single cells, a new method based on uniform and rotation invariant local binary patterns (LBP) algorithm for feature extraction was developed. This algorithm has advantages in gray-scale and rotation invariant and dimension reduction of the features. Each image was separated into 100 sub-images, features of each sub-image were extracted and then all features from the 100 sub-images were put together sequentially as a representation (feature vector) of the diffraction image. 2000 feature vectors were constructed for each image pair of a cell. A software tool based on support vector machine (SVM) with linear kernel and radial basis function (RBF) kernel were developed for classification. The training dataset consisted of samples of 2000 early apoptotic cells, 2000 viable cells and 1000 apoptotic/necrotic cells. In the test dataset, numbers of cell samples of these three sub- groups were 250, 250 and 80. After 10-fold cross validation, precision, Kappa statistic, mean absolute error and the area under the curve of the receiver operating characteristic (ROC area) were introduced to evaluate the prediction and classification model. Results Fig.1 shows typical diffraction images of K562 cells in different apoptotic stages. The Kappa statistic, mean absolute error and ROC area showed that the RBF kernel performed better than the linear kernel on the dataset (Table 1). Classification precision of 93.276% on the independent test dataset was obtained. Conclusion With the new method, stable and accurate classification of cells in different apoptotic stages can be achieved. EP-1894 On the possibility of estimating the radiosensitivity range in a cell mixture N. Stavreva 1 , P. Stavrev 1 , D. Penev 1 , A. Nahum 2 , R. Ruggieri 3 , D. Pessyanov 1 1 SRD Sofia University “St. Kliment Ohridski”, Scientific Research Dpt, Sofia, Bulgaria ; 2 Liverpool University, Physics Department, Liverpool, United Kingdom ; 3 ‘Sacrocuore – don Calabria’ Hospital, Department of Radiation Oncology, Negrar VR, Italy Purpose or Objective To expand a previous study [1] on tumours that are heterogeneous in cell radiosensitivity and on the observed differences between the values of the effective cell radiosensitivity extracted from fits to pseudo- experimental cell survival data and from fits to pseudo- experimental tumour control probability (TCP) data. Material and Methods

Conclusion A p-value of 0 renders the first version statistically unacceptable. This means that cell survival and the TCP experiments cannot be both described with a single a radiosensitivity value. This conclusion is confirmed by the high p-value (p=1) of the second fit, where two distinctly different best-fit values of α were obtained. А high value of α=0.35 Gy -1 from the fit to the cell-survival curve corresponds to the most sensitive sub-population and a relatively low value of α=0.12 Gy -1 from the fit to the TCP curve corresponds approximately to the most resistant cell subpopulation of the heterogeneous tumour. It should be noted that the best fit value of N o also corresponds to the number of cells of the most resistant component of the tumour. 1. Stavrev et al. Phys Med Biol 2015;60(15):N293-9. Supported by the Bulgarian National Science Fund under contract: DN 18/4 (10.12.2017).