S199
Clinical - Biomarkers of clinical response
ESTRO 2026
Material/Methods: This retrospective study included patients with histologically confirmed MLPS treated with preoperative RT (50 Gy in 25 fractions) between January 2015 and December 2024, without concurrent chemotherapy. Patients were eligible if pre-treatment MRI scans were available, including at least one T1- or T2-weighted sequence suitable for radiomic analysis. Tumour volumes were manually segmented in three dimensions using 3D Slicer, and 107 radiomic features were extracted from each sequence with PyRadiomics (v3.1.0). Image intensities were normalised using a reference region drawn over healthy tissue to reduce scanner-related variability. Pathological response was assessed on the surgical specimen and classified according to the percentage of tumour necrosis: good responders ( ≥ 90% necrosis) and poor responders (<90%). Statistical analyses included t-tests (or Welch’s t-test when appropriate) and Mann–Whitney tests to account for group imbalance. Bonferroni correction was applied for multiple testing, and no predictive models were developed to avoid overfitting. Results: Thirty-seven patients received neoadjuvant exclusive RT. Nineteen patients with evaluable T1-weighted MRI were analysed (10 poor responders, 9 good responders), and sixteen were included for T2- weighted analysis (11 poor, 5 good responders). Radiomic analysis revealed distinct pre-treatment imaging profiles between poor and good responders. In T1-weighted images, eleven features differed significantly (p < 0.05). Tumours with more regular geometry and higher internal homogeneity, reflected by features as shape_SurfaceVolumeRatio and glcm_Correlation, were associated with a better pathological response (AUC 0.81–0.86). In T2-weighted images, fifty-six features reached significance (p < 0.05). Good responders exhibited higher signal uniformity and lower tissue density, consistent with a more homogeneous myxoid matrix, while poor responders showed greater heterogeneity. The most discriminative features achieved AUC values up to 0.95. Conclusion: MRI-based radiomics can non-invasively characterise morphological and textural tumour properties associated with radiosensitivity in myxoid liposarcoma. Features reflecting geometric regularity and signal homogeneity correlated with ≥ 90% necrosis after preoperative RT, supporting their potential as quantitative imaging biomarkers for treatment planning and response prediction in MLPS. References: Qu G, Zhang C, Tian Z, Yao W. Diagnosis and Treatment of Myxoid Liposarcoma. Curr Treat Options Oncol. 2024 Oct;25(10):1289–96.Chung PWM, Deheshi BM, Ferguson PC, Wunder JS, Griffin AM, Catton CN, et
al. Radiosensitivity translates into excellent local control in extremity myxoid liposarcoma: a comparison with other soft tissue sarcomas. Cancer. 2009 Jul 15;115(14):3254–61.Sampath S, Schultheiss TE, Hitchcock YJ, Randall RL, Shrieve DC, Wong JYC. Preoperative versus postoperative radiotherapy in soft-tissue sarcoma: multi-institutional analysis of 821 patients. Int J Radiat Oncol Biol Phys. 2011 Oct 1;81(2):498–505. Keywords: Radiomics, Myxoid liposarcoma, Radiotherapy
Made with FlippingBook - Share PDF online