ESTRO 38 Abstract book

S1109 ESTRO 38

of the internal obturator muscles. Specifically, the increased signal intensity and homogeneous pattern at MRI2, compatible with an inflammatory status, was more evident in the region receiving high dose. Two years after the end of RT, a relatively high recovery was observed in the high dose region, indicating a partial resolution of the inflammation; however, the plateau found in the low dose region may indicates that this altered status could not be completely recovered. Other analyses at further time-points should be performed and correlations with genito-urinary toxicity scores should be investigated, considering the involvement of the pelvic floor muscles in the urinary dysfunctions. EP-2023 Predictive value of delta-radiomics features extracted from MR Images in image-guided liver SBRT N. Dogan 1 , D. Asher 1 , B. Farnia 1 , C. Ford 1 , F. Yang 1 , L. Portelance 1 , G. Simpson 1 1 University of Miami- Sylvester Comprehensive Cancer Center, Department of Radiation Oncology, Miami- Florida, USA Purpose or Objective With a projection of 600,000 deaths each year, liver cancer is a leading cause of cancer deaths worldwide and is very difficult to treat. A predictive model that could identify patients whose tumors are not responding to the treatment could be beneficial for improved personalized care. On-board MRI-guided radiotherapy which allows real-time tumor tracking presents a potential to utilize daily setup MR images to predict response and as well as safe dose escalation. The objective of this study was to determine the changes in image texture features (delta- radiomics) measured on daily low-field MR images and if delta-radiomics features could be used to assess treatment response and predict patient outcomes. Material and Methods We conducted a retrospective analysis of ten patients with liver cancer treated with MR-guided SBRT (27 Gy to 50 Gy in 3 to 5 fractions). Radiomics features extracted from set- up images were acquired with an onboard 0.35 T MRI. The predictive capabilities of texture features extracted from daily set-up images for patients with liver lesions treated with SBRT were assessed using delta-radiomics which track changes in texture features through treatment. Five classes of texture features containing 42 features were extracted from the gross tumor volume (GTV). Gray-level Size Zone Matrix (GLSZM), Neighborhood Gray-level Difference Matrix (NGLDM), Gray-level Co-occurrence Matrix (GLCOM), and Gray-level Run Length Matrix (GRLRM) based features were selected. After extraction of texture features, the difference in feature value between first and last fraction set-up images was calculated Patients were grouped into non-responders (three patients) who had local failure (disease progression) and responders (seven patients) who achieved local control. Patients were classified using follow-up PET images with modified RECIST criteria. The delta-radiomics features for each group were compared using Wilcoxon signed-rank test. Results A univariate analysis identified three delta-radiomics texture features ( p-value < 0.05) that were able to differentiate between responders and non-responders groups. The difference between dissimilarity (GLCOM), homogeneity (GLCOM), and busyness (GLNDM) features, was statically significant ( p-value < 0.05) between the two groups. Entropy (GLCOM), complexity (GLNDM), and gray- level non-uniformity (GLSZM) features approached significance ( p = 0.0874). No GRLRM or GLSZM features in the analysis were found to be significant. Conclusion The ability to continually inform physician decisions during treatment would significantly advance the move towards personalized care. Delta-radiomics has the potential to

Purpose or Objective To investigate radiation-induced alterations in internal obturator muscles in prostate cancer patients treated with external-beam radiotherapy (RT), considering T2w-MRI images, acquired before treatment and at one and two years of follow-up. Material and Methods T2-weighted MRI were acquired in fourteen patients before RT (MRI1), about 12 months (MRI2) and 24 months (MRI3) of follow-up. Images were corrected for bias field inhomogeneities, normalized and spatially registered to automatically propagate obturators contours, which were manually delineated on MRI1. The following textural features were extracted: histogram-based indices (mean intensity, variance, 95 th percentile, entropy, skewness, kurtosis), GLCM (Grey- Level Co-occurrence Matrix)-based indices (energy, correlation, homogeneity, entropy, contrast, dissimilarity), NGTDM (Neighborhood Grey-Tone Different Matrix)-based indices (coarseness, contrast, busyness, complexity, strength) and fractal dimension (FD). The dose map associated with the planning CT was deformed on each MRI and two regions were identified within the obturator: a high dose region (receiving >55 Gy) and a low dose region (receiving <55 Gy). To assess significant changes in internal obturator muscles, a comparison of the parameters extracted at each time point in the two regions was carried out by repeated-measure ANOVA and post-hoc t-test, with significant p-value < 0.05, after Bonferroni correction for multiple comparisons. Results For several features (such as mean value, variance, 95 th percentile, GLCM homogeneity and correlation, FD) it was found a significant increase or decrease at MRI2 and a gradual recovery at MRI3, in some cases still significantly different from MRI1 (see Figure 1).

Looking at regional results, different trends were found for the two regions. The high dose region presented highly different values at MRI2 with respect to MRI1 and a rapid but incomplete recovery at MRI3. The low dose region presented lower and less significant differences at MRI2, with respect to the high dose region, followed by a plateau at MRI3 (see Figure 2).

Conclusion In patients who underwent RT for prostate cancer treatment, a dose-dependent behaviour was observed regarding changes in T2w-MRI signal intensity and texture