S2464
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
Material/Methods: We retrospectively evaluated 26 patients with
biomarker to improve the detection and monitoring of RT-induced bone injury. References: 1. Watson EE, Hueniken K, Lee J, et al. Development and Standardization of an Osteoradionecrosis Classification System in Head and Neck Cancer: Implementation of a Risk-Based Model. J Clin Oncol. 2024;42(16):1922-1933. doi:10.1200/JCO.23.01951.2. Eley KA, Mcintyre AG, Watt-Smith SR, Golding SJ. “Black bone” MRI: a partial flip angle technique for radiation reduction in craniofacial imaging. Br J Radiol. 2012;85(1011):272-278. doi:10.1259/bjr/95110289.3. Mackay K, Bernstein D, Glocker B, Kamnitsas K, Taylor A. A Review of the Metrics Used to Assess Auto- Contouring Systems in Radiotherapy. Clin Oncol. 2023;35(6):354-369. doi:10.1016/j.clon.2023.01.016. Keywords: Head and neck, osteoradionecrosis, biomarker Predicting high-grade glioma response: Comparing 1D, 2D, and 3D neural networks with a mechanistic model using a novel data assimilation pipeline. Hugo JM Miniere 1 , David A Hormuth 2 , Maguy Farhat 1 , Bikash Panthi 1 , Holly Langshaw 1 , Mihir D Shanker 1 , Wasif Talpur 1 , Sara Thrower 3 , Jodi Goldman 1 , Caroline Chung 1 1 Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, USA. 2 Oden Institute for Computational Engineering and Sciences, The University of Texas at Austin, Austin, USA. 3 Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, USA Purpose/Objective: High-grade gliomas are aggressive tumors that can demonstrate changes during a course of chemoradiation (CRT) [1]. There is growing ability for MR-based adaptive radiation (RT), including the use of MR-linac for high frequency assessment and Digital Poster Highlight 3505 adaptation [2]. The aim of this study is to evaluate a novel computational pipeline that assimilates newly gathered MRI data to perform AI-based sequential predictions of spatiotemporal tumor growth during CRT to inform treatment planning. Material/Methods: Weekly T1-weighted, T2-FLAIR, and diffusion-weighted MRI data were collected as part of prospective trial evaluating adaptive CRT for high-grade glioma. Neural networks were trained using a leave-one-out approach (ntraining = 20, ntesting = 1), in which cellularity maps obtained from baseline through visit N were used to predict cellularity at visit N+1 throughout the entire course of radiation treatment. We investigated three different data and network structures: (1) voxel-wise
squamous cell carcinoma. Among these, 6 had active ORN. Patients were scanned on a CT and 1.5-Tesla MR simulation scanner at our institution between 2017- 2023. All scans occurred within 4 weeks of each other to minimize anatomical variability. Mandibles were manually segmented on CT, T2-weighted (T2W) MRI, and black bone (BB) MRI; all images were co-registered prior to analysis. Image segmentation evaluation metrics, including dice similarity coefficient (DSC), mean distance to agreement (MDA), and Hausdorff distance (HD)3, quantitatively compared spatial differences between each MRI sequence, using CT as reference standard. A Wilcoxon-signed rank test with a significance value of p ≤ 0.05 was used to assess for statistical significance between T2W and BB images. Results: When evaluating 26 patients, all comparison metrics showed a significant difference between BB and T2W MRI when compared against CT (Figure 1), indicating a distinction between mandible contour and image modality. BB had significantly higher DSC (median=0.886, IQR=0.0508) than T2W (median=0.818, IQR=0.0832; p<0.0001). BB had a significantly lower MDA (median=0.0659mm, IQR=0.0288mm) than T2W (median=0.117mm, IQR=0.0691mm; p<0.0001). Furthermore, BB had a significantly lower HD (median=0.618mm, IQR=0.214mm) than T2W (median=1.73mm, IQR=1.13mm; p<0.0001). When evaluating only ORN patients, the same relationships were upheld. BB had significantly higher DSC (median=0.920, IQR=0.0131) than T2W (median=0.863, IQR=0.0129; p=0.02). BB had a significantly lower MDA (median=0.0464mm, IQR=0.0140mm) than T2W (median=0.0842mm, IQR=0.00488mm; p=0.02). Finally, BB had a significantly lower HD (median=0.433mm, IQR=0.220mm) than T2W (median=0.814mm, IQR=0.200mm; p=0.02). Conclusion:
This morphometric analysis demonstrates black bone MRI as a non-ionizing alternative with spatial performance comparable to CT. Integrating black bone into follow-up MR imaging can lead to earlier, more objective ORN diagnosis. These findings underscore the potential to develop black bone into an imaging
Made with FlippingBook - Share PDF online