ESTRO 2025 - Abstract Book

S3867

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Keywords: Head and Neck Cancer, Overall Survival Prediction

References: 1.Budach, V. & Tinhofer, I. Novel prognostic clinical factors and biomarkers for outcome prediction in head and neck cancer: a systematic review. The Lancet Oncol. 20, e313–e326 (2019). Publisher: Elsevier. 2. Li, Y., Wehbe, R. M., Ahmad, F. S., Wang, H. & Luo, Y. Clinical-Longformer and Clinical-BigBird: Transformers for long clinical sequences (2022). Number: arXiv:2201.11838 arXiv:2201.11838 [cs]. 3. Pölsterl, S. scikit-survival: a library for time-to-event analysis built on top of scikit-learn. J. Mach. Learn. Res. 21, 1– 6 (2020).

4580

Digital Poster Perfusion MRI-radiomics for non-invasive differentiation of tumour progression and radionecrosis in stereotactic radiosurgery patients Michael Maddalena 1 , Adam Farag 2 , Paula Alcaide-Leon 2 , David B Shultz 3,4 , Catherine Coolens 1,3 1 Medical Biophysics, University of Toronto, Toronto, Canada. 2 Medical Imaging, University Health Network, Toronto, Canada. 3 Radiation Oncology, University of Toronto, Toronto, Canada. 4 Institute of Medical Sciences, University of Toronto, Toronto, Canada Purpose/Objective: Despite excellent local control of brain metastasis patients treated with stereotactic radiosurgery (SRS), potential tumour progression (TP) is observed during follow-up in up to 20% of cases 1 . Lesions may also contain radionecrosis (RN), a by-product of SRS that appears indistinguishable from TP via MRI but requires a distinct treatment regimen 2 . Proper diagnostic protocols are crucial for prolonging patient survival post-SRS, as up to 20% of all cancer patients develop brain metastases, a condition currently associated with a dismal 24-month overall survival rate of 8.1% 3 . Currently, invasive post-surgical histopathology remains the only gold-standard confirmation of TP/RN. Thus, novel imaging protocols with strong diagnostic accuracy are required to improve stratification while minimizing harm. However, the historic rate of image-based TP/RN differentiation accuracy is a modest 54%, impeding the widespread adoption of image-based protocols in clinical practice due to suboptimal performance 4 . To address this gap, our study aims to establish a perfusion MRI classification protocol to distinguish tumour progression from radionecrosis in patient lesions post-SRS at a ≥80% diagnostic AUC/sensitivity/specificity threshold across all implemented classification methods. Material/Methods: Patients with confirmed TP/RN via histopathology (n=9) have been enrolled in an ongoing clinical trial. Internal datasets from recent brain metastasis-SRS studies were included alongside study data to enhance classifier robustness (n=12). Standard MR sequences, including T1-weighted post-contrast and DWI-apparent diffusion coefficient (ADC), were acquired alongside perfusion-weighted sequences, such as Dynamic Susceptibility Contrast (DSC) and Dynamic Contrast Enhanced (DCE). Lesion contours were translated to ROI masks and interpolated to divide the lesion volume into three non-overlapping regions: the lesion core, the inner lesion periphery, and the lesion edge. Radiomic features were extracted per-volume for all sequences and were classified via a Random Forest ensemble model, which was evaluated using the leave-one-out cross-validation method. ROC-AUC, sensitivity and specificity scores were reported to compare the performance of each sequence for TP/RN differentiation. Results: Classification of perfusion-weighted parameters yielded results superior to the target threshold (DCE-k trans parameter: ROC-AUC = 0.874, specificity = 0.833, sensitivity = 0.914), while standard MR sequences yielded modest results (T1-weighted post-contrast: ROC-AUC = 0.594, specificity = 0.4, sensitivity = 0.787; ADC ROC-AUC = 0.665, specificity = 0.5, sensitivity = 0.830).