MRI-based radiomic analysis for grading myxoid liposarcoma: a multisequence retrospective study

Objectives: This study aimed to identify quantitative MRI features through radiomic analysis and to develop predictive models for determining the histological grade of myxoid liposarcoma (MLS). Materials and methods: This retrospective single-center study included 57 patients with histologically confirmed MLS (30 low-grade, 27 high-grade). Tumors were segmented and 107 radiomic features were extracted from T1-weighted imaging (WI), T2-WI, short tau inversion recovery (STIR), apparent diffusion coefficient (ADC) maps, and contrast-enhanced (CE) images with and without fat saturation (FS). Features showing statistical significance (p < 0.05) were selected and used to develop predictive models, whose performance was assessed using cross-validation and reported as area under the curve (AUC). Results: Mean age was 51.6 ± 14.7 years (32 men, 25 women). Radiomic analysis identified three significant features for T1-WI and STIR and 19 for T2-WI. For CE-T1-WI, CE-T1-FS-WI, and CE-3D, four, six, and three features were significant, respectively. Models based on T2-WI and CE-3D achieved the highest performance (AUC up to 0.88). Additional models trained exclusively on institutional T1-WI and T2-WI showed reduced performance on external validation, although AUCs improved when applied to patients scanned with the same vendor. Conclusion: Radiomic analysis of pre-treatment MRI shows promising results in predicting histological grade of MLS. This study is novel in addressing grading rather than diagnosis alone, a distinction with clear clinical relevance for treatment planning and prognostic assessment. In particular, models based on T2-WI may complement conventional imaging and histopathology by providing whole-tumor quantitative grading, while multicentric validation is required for clinical application.