ESTRO 2022 - Abstract Book

S404

Abstract book

ESTRO 2022

Figure 1. Analysis flowchart. Step 1, The radiomics and dosiomics features of the lung tissue region were extracted. Step 2, 1000 unique bootstrap samples were taken from all samples, features were selected by correlation, least absolute shrinkage (LASSO) embedded with logistic regression (LR) and Akaike information criterion (AIC) for modeling. Step 3, The model performance was evaluated using discrimination and calibration. Step 4, Clinical applications were evaluated using nomogram and decision curves. Results A model built by integrating all of R-score, D-score and C-score had the best discriminative ability with area under the curves (AUCs) of 0.793,0.774 and 0.780 in the training set, bootstrapping set and external test set, respectively. The results of the calibration curve and decision curve analysis showed that the final model of the nomogram has potential for future clinical application. Table 1 Discrimination ability of different models according to area under the curve (AUC) with the range provided between parentheses. Model Train Validation by bootstrapping Testing Radiomics 0.676 (0.628-0.777) 0.619 (0.541-0.639) 0.636 (0.585-0.688) Dosiomics 0.728 (0.621-0.814) 0.687 (0.639-0.704) 0.630 (0.594-0.667) Clinical 0.664 (0.548-0.784) 0.654 (0.529-0.662) 0.646 (0.625-0.676) R-score+D-score 0.735 (0.636-0.829) 0.728 (0.692-0.731) 0.679 (0.641-0.718) R-score+C-score 0.717 (0.594-0.816) 0.701 (0.636-0.709) 0.723 (0.692-0.756) D-score+C-score 0.770 (0.641-0.873) 0.755 (0.724-0.771) 0.770 (0.726-0.808) R-score+D-score+C-score 0.793 (0.667-0.891) 0.774 (0.739-0.781) 0.780 (0.744-0.821)

Conclusion Radiomics and dosiomics features have potential to assist with the prediction of RP, and the combination of radiomics, dosiomics and clinical parameters led to the best prognostic model in the present study.

OC-0459 Tumor Control Probability following Radiosurgery of Brain Metastases With and Without Retreatment

M. Sharma 1 , M.T. Milano 2 , M. Cummings 2 , I.E. Naqa 3

1 UCSF, Radiation Oncology, San Francisco, USA; 2 University of Rochester Medical Center, Radiation Oncology, Rochester, USA; 3 Moffitt Cancer Center, Radiation Oncology, Florida, USA Purpose or Objective Tumor control probability (TCP) models were developed to quantify the relationship between radiation dose and local control after single-fraction stereotactic radiosurgery (SRS) for brain metastasis with and without retreatment. Materials and Methods Patients treated with single-fraction LINAC-based SRS at a single institution from 2003-2016 were used to model TCP. The non-retreatment cohort included 173 patients with 493 metastatic lesions while the retreatment cohort included 25 patients with 50 metastatic lesions. For retreatment, the patients were usually prescribed the same tumor dose irrespective of the dose received in the initial treatment. Following the recent HyTEC methodology the dose to 99% of each lesion’s planning-