MRI features outperform histologic grade for outcome prediction in translocation-driven enriched soft-tissue sarcoma Authors: Han-Ying Lin, Ching-Lan Wu, Chih-Hsueh Chen, Chueh-Chuan Yen, Chao-Ming Chen, Chien-An Liu, Po-Kuei Wu, Hung-Ta Wu, Chung-Jung Lin Published: 2026 ABSTRACT AbstractBackgroundMost MR-based prognostic series in soft-tissue sarcoma (STS) were derived in cohorts dominated by complex-karyotype histologies; whether conventional MR descriptors retain prognostic value in translocation-driven enriched populations is unclear. The purpose of our study is to identify preoperative MR features independently associated with overall survival (OS), local recurrence-free survival (LRFS), and metastasis-free survival (MFS) in a translocation-driven enriched soft-tissue sarcoma cohort.MethodsIn this retrospective single-center study, consecutive adults with histologically confirmed STS who underwent contrast-enhanced MRI before treatment between January 2008 and December 2024 were evaluated. Two radiologists blinded to outcome independently scored 13 prespecified MR features. Cox regression adjusted for age, sex, maximum tumor diameter, and Fédération Nationale des Centres de Lutte Contre le Cancer (FNCLCC) grade was used to estimate hazard ratios (HRs); a sensitivity analysis added translocation-driven histology.ResultsA total of 158 patients (mean age, 48 years ± 19 [SD]; 84 men) were evaluated; 61 of 152 classifiable patients (40%) had translocation-driven tumors. After adjustment for age, sex, and maximum tumor diameter, heterogeneous gadolinium enhancement was independently associated with shorter OS (HR, 3.9 [95% CI: 1.4, 10.7]; P = .008). Peritumoral edema was independently associated with shorter OS (HR, 2.0 [95% CI: 1.0, 3.7]; P = .04) and was the single independent MR predictor of LRFS (HR, 2.3 [95% CI: 1.2, 4.5]; P = .01). Both associations persisted after adjustment for translocation-driven histology. A 13-feature MR panel did not discriminate FNCLCC grade 3 (area under the curve [AUC], 0.59).ConclusionPeritumoral edema and heterogeneous gadolinium enhancement were independent, complementary MR predictors of outcome in translocation-driven enriched soft-tissue sarcoma, outperforming FNCLCC grade and molecular histologic class. These findings require prospective multicenter validation before clinical application.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s40644-026-01080-2.Keywords: Soft-tissue sarcoma, MRI, Peritumoral edema, Gadolinium enhancement, Survival, Local recurrence, Translocation-driven sarcoma, FNCLCC grade Background Soft-tissue sarcomas (STS) are a biologically heterogeneous group of mesenchymal malignancies that account for fewer than 1% of adult cancers but contribute disproportionately to morbidity because they are frequently diagnosed at a locally advanced stage and recur despite aggressive multimodal therapy [1]. Preoperative MRI is central to defining local extent and guiding surgical planning [2–4], yet only a subset of conventional MR descriptors has been shown to carry prognostic weight [5–7]. Most published MR-prognostic series were derived in cohorts dominated by older adults and complex-karyotype histologies [6, 8, 9]. Translocation-driven subtypes (for example synovial sarcoma, myxoid liposarcoma, alveolar soft-part sarcoma, clear-cell sarcoma, dermatofibrosarcoma protuberans, and low-grade fibromyxoid sarcoma, each defined by a recurrent gene fusion) — which can behave differently biologically — are under-represented in these cohorts, and the MR descriptors that retain independent prognostic value in translocation-driven enriched populations have not been systematically tested [10–12]. We therefore re-examined a prespecified panel of 13 conventional MR features in a single-center cohort that included a higher-than-usual proportion of translocation-driven sarcomas. We deliberately analyzed the full sarcoma cohort, enriched for but not restricted to translocation-driven subtypes, so that the resulting descriptors could be applied prospectively at the time of MRI — before histologic and molecular classification are known — and would remain generalizable to unselected sarcoma populations. Our aim was to identify descriptors that retain independent prognostic value after adjustment for age, sex, tumor size, and Fédération Nationale des Centres de Lutte Contre le Cancer (FNCLCC) grade. We hypothesized that two binary visual descriptors — peritumoral edema and heterogeneous gadolinium enhancement — would independently predict overall survival and local recurrence-free survival and that this signal would be independent of histologic grade and provide information that grade does not capture. Methods Study design and cohort This retrospective single-institution study was approved by our institutional review board, which waived the requirement for informed consent. Consecutive adults (≥ 18 years) with histologically confirmed soft-tissue sarcoma who underwent contrast-enhanced MRI before any treatment between January 2008 and December 2024 were eligible. Exclusion criteria were head-and-neck, visceral, or retroperitoneal location; Ewing sarcoma; prior resection or chemoradiation; suboptimal imaging quality (e.g., lack of contrast-enhanced sequences); and missing survival data. After applying these criteria, 158 patients remained for analysis (Fig. 1). Clinical variables — age, sex, maximum tumor diameter, anatomic site, histologic subtype, FNCLCC grade, resection margin status, and neoadjuvant and adjuvant therapies — were abstracted from the electronic medical record. All patients were treated per contemporary ESMO and NCCN guidelines [13–16]. This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. Patient selection and analysis flow MR protocol and image analysis MR examinations were performed on 1.5-T or 3-T scanners, at our institution or at outside facilities, before any treatment. Each examination included at minimum axial T1-weighted, T2-weighted with or without fat suppression, coronal or sagittal T2-weighted with or without fat suppression, and post-gadolinium T1-weighted sequences. Studies were reviewed on a dedicated workstation by two radiologists (reader 1, 15 years of musculoskeletal MRI experience; reader 2, 6 years), blinded to clinical outcome and histologic grade. Readers independently scored 13 prespecified MR features [8, 17]: T1 signal heterogeneity, T2 signal heterogeneity, heterogeneous gadolinium enhancement (each defined as visually discernible spatial variation in signal intensity involving more than 50% of the tumor cross-section on a representative slice), peritumoral enhancement, peritumoral edema, intratumoral necrosis, intratumoral hemorrhage, tumor margin (well-defined vs. infiltrative) in T1 and T2, neurovascular encasement, bone invasion, nodal metastasis, and tail sign. The 13 features were prespecified from descriptors previously reported to carry prognostic value in soft-tissue sarcoma, so that any prognostic signal identified could be translated into practice without specialized acquisition [6–7, 9]. Each feature except tumor margin was scored as present or absent. Disagreements were resolved by consensus. Reader 1 rescored all examinations after a minimum interval of 8 weeks between the two scoring sessions, to estimate intra-reader agreement and reduce the chance that the reader recalled the first reading. Inter- and intra-reader agreement was summarized with the Cohen κ overall and per feature; per-feature values are reported in Additional Table A1. Maximum tumor diameter was measured on the coronal or sagittal plane best capturing the long axis. Histology classification Histologic diagnoses were established on surgical specimens by two dedicated bone-and-soft-tissue pathologists using the 2020 WHO classification [5], with discordant cases resolved by joint review (concordance, 149 of 152; 98%); six cases with uncertain molecular classification were excluded from histology-adjusted analyses. Whenever available, diagnoses were confirmed via molecular testing, such as fluorescence in situ hybridization or reverse transcription-polymerase chain reaction, identifying characteristic fusion transcripts [18]. Tumors were classified as translocation-driven when the diagnosis corresponded to a sarcoma subtype defined by a recurrent, characteristic gene fusion — for example synovial sarcoma (SS18::SSX), myxoid liposarcoma (FUS::DDIT3), alveolar soft-part sarcoma (ASPSCR1::TFE3), clear-cell sarcoma (EWSR1::ATF1), dermatofibrosarcoma protuberans (COL1A1::PDGFB), or low-grade fibromyxoid sarcoma (FUS::CREB3L2) — confirmed by fluorescence in situ hybridization or reverse transcription–polymerase chain reaction whenever molecular testing was available [19, 20]; all remaining subtypes, characterized by complex, non-recurrent genomic alterations rather than a single driver fusion (e.g., undifferentiated pleomorphic sarcoma and spindle-cell sarcoma not otherwise specified), were classified as complex karyotype [10, 21]. Outcomes Three prespecified endpoints were overall survival (OS; time from MRI to all-cause death), local recurrence-free survival (LRFS; time to biopsy-proven or imaging-confirmed recurrence at the primary site), and metastasis-free survival (MFS; time to first distant metastasis). Patients without events were censored at the date of last clinical contact. Statistical analysis Continuous variables are summarized as mean ± SD or median (IQR); categorical variables as counts and percentages. Between-group differences were assessed with the two-sample t test, Wilcoxon rank-sum test, χ² test, or Fisher exact test, as appropriate. Kaplan–Meier estimates were compared with the log-rank test. Univariable and multivariable Cox proportional hazards regression was used to estimate hazard ratios (HRs) with 95% CIs for each of the 13 prespecified MR features. The primary multivariable model adjusted for age, sex, maximum tumor diameter, and FNCLCC grade; ridge penalization (λ = 0.01) was applied to stabilize estimates. A sensitivity analysis added translocation-driven histology as a fifth covariate. Translocation-driven histology was modeled as a covariate rather than as an inclusion criterion so that the prognostic value of the MR features could be assessed across the full sarcoma spectrum and tested for independence from molecular class. Proportional hazards assumptions were assessed with Schoenfeld residuals; no violation was detected for the primary multivariable covariates (all global tests P > .10). To test whether heterogeneous gadolinium enhancement acted as a grade surrogate, we fit a logistic regression of FNCLCC grade 3 (vs. grade 1–2) on the 13 MR features with age, sex, and maximum diameter; discrimination was summarized as area under the receiver operating characteristic curve (AUC). Per the prespecified Statistical Analysis Plan, peritumoral edema and peritumoral enhancement were not entered simultaneously because of collinearity; peritumoral edema was the primary peritumoral-domain covariate, and peritumoral enhancement was tested in a drop-one sensitivity model. Median follow-up was estimated using the reverse Kaplan–Meier method [22]. All P values are two-sided; P < .05 was considered significant. Analyses were performed with Python 3.11.8 (lifelines 0.27.8, scikit-learn 1.4.1, pandas 2.2.0, NumPy 1.26.3). Results Patient characteristics The study included 158 patients (mean age, 48 years ± 19 [SD]; 84 men [53%]). Median tumor diameter was 77 mm (IQR: 48, 114), and 124 of 158 patients (79%) had FNCLCC grade 3 disease. The most common histologic subtypes were synovial sarcoma (34 of 158 [22%]), undifferentiated pleomorphic sarcoma (30 of 158 [19%]), and spindle-cell sarcoma not otherwise specified (21 of 158 [13%]). Molecular status was classifiable in 152 patients, of whom 61 (40%) had translocation-driven tumors — predominantly synovial sarcoma (34 of 61 [56%]) and myxoid liposarcoma (13 of 61 [21%]); the full subtype breakdown appears in Table 1. Heterogeneous gadolinium enhancement was present in 118 of 158 patients (75%) and peritumoral edema in 105 of 158 (66%). Patients with heterogeneous enhancement were older (mean, 51 vs. 41 years; P = .015), had larger tumors (median, 86 vs. 52 mm; P = .001), and were more often men (58% vs. 38%; P = .035); those with peritumoral edema were also older (51 vs. 44 years; P = .04) and had larger tumors (86 vs. 69 mm; P = .01). FNCLCC grade 3 proportion did not differ by either feature (Table 1). Table 1Baseline characteristics, stratified by heterogeneous gadolinium enhancementVariableAll (N = 158)Homogeneous Gd(N = 40)Heterogeneous Gd(N = 118)P valueAge, years (mean ± SD)48.4 ± 19.041.4 ± 21.550.7 ± 17.5 0.015* Male sex84 (53.2%)15 (37.5%)69 (58.5%) 0.035* Max diameter, mm (median, IQR)77.0 (48.0-113.8)52.5 (33.8–86.2)86.0 (57.2-123.8) 0.001* > 50 mm115 (72.8%)21 (52.5%)94 (79.7%)> 100 mm48 (30.4%)7 (17.5%)41 (34.7%) FNCLCC grade 0.683Grade 11 (0.6%)1 (2.5%)0 (0.0%)Grade 232 (20.3%)6 (15.0%)26 (22.0%)Grade 3124 (78.5%)33 (82.5%)91 (77.1%)Translocation-driven histology61/152 (40.1%)25/40 (62.5%)36/112 (32.1%) 0.002* MR features (present) T1 heterogeneous92 (58.2%)16 (40.0%)76 (64.4%) 0.012* T2 heterogeneous114 (72.2%)18 (45.0%)96 (81.4%) < 0.001† Peritumoral edema105 (66.5%)21 (52.5%)84 (71.2%)0.053Peritumoral enhancement101 (63.9%)21 (52.5%)80 (67.8%)0.121Intratumoral necrosis73 (46.2%)5 (12.5%)68 (57.6%) < 0.001† Intratumoral hemorrhage35 (22.2%)7 (17.5%)28 (23.7%)0.549Ill-defined margin on T143 (27.2%)10 (25.0%)33 (28.0%)0.84Ill-defined margin on T243 (27.2%)10 (25.0%)33 (28.0%)0.84Tail sign40 (25.3%)12 (30.0%)28 (23.7%)0.563Bone invasion48 (30.4%)6 (15.0%)42 (35.6%) 0.025* Nodal metastasis10 (6.3%)1 (2.5%)9 (7.6%)0.454Vascular/nerve invasion35 (22.2%)4 (10.0%)31 (26.3%) 0.046* Margin status Margin not free (documented)4 (2.5%)0 (0.0%)4 (3.4%) Outcomes Deaths58 (36.7%)6 (15.0%)52 (44.1%) 0.002* Local recurrences53 (33.5%)11 (27.5%)42 (35.6%)0.457Distant metastases73 (46.2%)12 (30.0%)61 (51.7%) 0.028* Observation time, months (median, IQR)‡38.8 (17.0-76.1)66.4 (46.1–93.7)31.3 (11.8–62.9)—Abbreviations: Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; IQR, interquartile range; SD, standard deviation. Significance markers: * p < .05, †p < .001. Note. Gd-restricted subcohort (cases with valid heterogeneous gadolinium enhancement scoring, N = 158). Continuous variables compared with Mann–Whitney U; categorical with χ² or Fisher exact test. FNCLCC grade was assignable in 157 of 158 patients (1 case with ungradable core biopsy). Translocation-driven histology is classifiable in 152/158 cases (6 with uncertain histology class excluded). ‡ Observation time denotes the median time from MRI to death or last clinical contact. It is not the reverse Kaplan–Meier estimator of median follow-up (68 months; see Results). Because observation time is truncated by deaths, between-group comparison mechanically reflects the survival difference and is not reported Baseline characteristics, stratified by heterogeneous gadolinium enhancement Abbreviations: Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer; IQR, interquartile range; SD, standard deviation. Significance markers: * p < .05, †p < .001. Note. Gd-restricted subcohort (cases with valid heterogeneous gadolinium enhancement scoring, N = 158). Continuous variables compared with Mann–Whitney U; categorical with χ² or Fisher exact test. FNCLCC grade was assignable in 157 of 158 patients (1 case with ungradable core biopsy). Translocation-driven histology is classifiable in 152/158 cases (6 with uncertain histology class excluded). ‡ Observation time denotes the median time from MRI to death or last clinical contact. It is not the reverse Kaplan–Meier estimator of median follow-up (68 months; see Results). Because observation time is truncated by deaths, between-group comparison mechanically reflects the survival difference and is not reported Follow-up and survival Median reverse Kaplan–Meier follow-up was 68 months for OS, 52 months for LRFS, and 65 months for MFS. During follow-up, 58 patients died, 53 developed local recurrence, and 73 developed distant metastasis. Kaplan–Meier curves stratified by the four MRI phenotypes defined jointly by peritumoral edema and heterogeneous gadolinium enhancement separated progressively for both endpoints (Fig. 2). For overall survival (Fig. 2a), 5-year estimates were 95% in patients with neither feature, 81% with peritumoral edema only, 64% with heterogeneous enhancement only, and 42% with both features (log-rank P < .001). For local recurrence–free survival (Fig. 2b), the 5-year estimates were 87%, 58%, 75%, and 48%, respectively (log-rank P = .005). The two single-feature phenotypes ranked discordantly across endpoints, consistent with peritumoral edema indexing local aggressiveness and heterogeneous enhancement indexing systemic risk. OS estimates beyond 120 months reflect tail-of-curve extrapolation (fewer than 20% of patients remained at risk). Kaplan–Meier estimates of (a) overall survival and (b) local recurrence–free survival in 158 adults with soft-tissue sarcoma, stratified by four MRI phenotypes defined jointly by peritumoral edema (PE) and heterogeneous gadolinium enhancement (HGE): neither feature (green; n = 19), peritumoral edema only (orange; n = 21), heterogeneous enhancement only (purple; n = 34), and both features (red; n = 84) MR features and overall survival Inter-reader agreement for the 13 features was substantial (κ = 0.86); intra-reader agreement was 0.91 for reader 1 and 0.84 for reader 2 (Additional Table A1). At univariable Cox analysis, the strongest MR predictors of shorter OS were heterogeneous gadolinium enhancement (HR, 4.4 [95% CI: 2.0, 9.8]; P < .001), vascular–nerve invasion (HR, 2.4; P = .002), intratumoral necrosis (HR, 2.2; P = .004), peritumoral edema (HR, 2.1; P = .02), and T1/T2 heterogeneity (HR ≈ 2.0; both P < .05). Maximum diameter (HR per 10-mm increment, 1.03 [95% CI: 1.01, 1.05]; P < .001), male sex (HR, 2.1; P = .008), and age (HR per year, 1.02; P = .008) were also significant; FNCLCC grade 3 was not (HR, 1.4; P = .38). In the primary multivariable model (n = 151; 54 events; C-index, 0.77), heterogeneous enhancement (HR, 3.9 [95% CI: 1.4, 10.7]; P = .008) and peritumoral edema (HR, 2.0 [95% CI: 1.0, 3.7]; P = .04) remained independently associated with shorter OS (Table 2). Table 2Univariable and multivariable Cox regression of MR features in predicting overall survivalVariableUnivariableHR (95% CI)P valueMultivariableHR (95% CI)P valueAge (per year)1.02 (1.00–1.03) 0.008* 1.02 (1.01–1.04) 0.010* Male sex2.10 (1.21–3.63) 0.008* 2.03 (1.11–3.70) 0.021* Maximum diameter (per 10 mm)1.03 (1.01–1.05) < 0.001† 1.04 (0.99–1.06)0.076FNCLCC grade 31.35 (0.69–2.65)0.3791.55 (0.74–3.26)0.248Translocation-driven histology0.63 (0.36–1.09)0.101——T1 heterogeneous signal2.01 (1.14–3.54) 0.015* 1.98 (0.99–3.99)0.055T2 heterogeneous signal2.05 (1.06–3.99) 0.034* 0.68 (0.30–1.54)0.359Peritumoral edema2.12 (1.16–3.89) 0.015* 1.95 (1.03–3.71) 0.042* Heterogeneous Gd enhancement 4.39 (1.97–9.81) < 0.001† 3.90 (1.42–10.68) 0.008* Peritumoral enhancement1.39 (0.80–2.41)0.239——Intratumoral necrosis2.17 (1.27–3.69) 0.004* 1.17 (0.65–2.12)0.600Intratumoral hemorrhage0.88 (0.48–1.63)0.689——Ill-defined margin on T11.40 (0.51–3.85)0.518——Ill-defined margin on T21.36 (0.49–3.75)0.556——Tail sign0.90 (0.49–1.66)0.736——Bone invasion1.75 (1.03–2.96) 0.038* 1.34 (0.75–2.38)0.325Nodal metastasis1.32 (0.48–3.62)0.586——Vascular/nerve invasion2.38 (1.38–4.12) 0.002* 1.59 (0.86–2.93)0.141Abbreviations: HR, hazard ratio; CI, confidence interval; OS, overall survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001.Primary multivariable Cox model: N = 151, events = 54, C-index = 0.767. Forced covariates: Age, Sex, Maximum diameter and FNCLCC Grade 3. Additional MR features entered if univariable p < .10. Ridge penalty λ = 0.01 Univariable and multivariable Cox regression of MR features in predicting overall survival Abbreviations: HR, hazard ratio; CI, confidence interval; OS, overall survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001.Primary multivariable Cox model: N = 151, events = 54, C-index = 0.767. Forced covariates: Age, Sex, Maximum diameter and FNCLCC Grade 3. Additional MR features entered if univariable p < .10. Ridge penalty λ = 0.01 MR features and local recurrence–free survival At univariable analysis, the strongest LRFS predictors were peritumoral edema (HR, 2.5 [95% CI: 1.3, 4.8]; P = .007) and peritumoral enhancement (HR, 2.4 [95% CI: 1.2, 4.5]; P = .009) — correlated manifestations of the peritumoral microenvironment (φ, 0.48; variance inflation factor, 1.5). Heterogeneous enhancement was a weaker univariable signal (HR, 2.0 [95% CI: 1.0, 3.9]; P = .06), and FNCLCC grade was not significant. In the prespecified multivariable model (n = 154; 50 events; C-index, 0.65), peritumoral edema was the only independent predictor (HR, 2.3 [95% CI: 1.2, 4.5]; P = .01); heterogeneous enhancement did not reach independent significance (HR, 1.7 [95% CI: 0.8, 3.4]; P = .14). A supplementary model substituting peritumoral enhancement for peritumoral edema yielded a comparable independent effect (HR, 2.2 [95% CI: 1.1, 4.2]; P = .02), indicating that the peritumoral-domain signal is robust to feature choice (Table 3). Table 3Univariable and multivariable Cox regression of MR features in predicting local recurrence free survivalVariableUnivariableHR (95% CI)P valueMultivariableHR (95% CI)P valueAge (per year)1.01 (1.00–1.02)0.2021.00 (0.98–1.02)0.985Male sex1.20 (0.69–2.08)0.5181.12 (0.62–1.99)0.713Maximum diameter (per 10 mm)1.00 (0.97–1.04)0.8341.00 (0.95–1.04)0.999FNCLCC grade 30.70 (0.38–1.29)0.2460.68 (0.35–1.31)0.247Translocation-driven histology0.88 (0.50–1.54)0.646——T1 heterogeneous signal0.82 (0.47–1.41)0.470——T2 heterogeneous signal1.26 (0.68–2.36)0.461——Peritumoral edema 2.47 (1.28–4.78) 0.007* 2.32 (1.19–4.52) 0.013* Heterogeneous Gd enhancement 1.95 (0.97–3.90)0.0601.68 (0.84–3.37)0.140Peritumoral enhancement2.37 (1.24–4.54) 0.009* — (excl., SAP)—Intratumoral necrosis1.19 (0.68–2.08)0.552——Intratumoral hemorrhage0.54 (0.26–1.13)0.101——Ill-defined margin on T12.45 (0.63–9.52)0.195——Ill-defined margin on T22.48 (0.64–9.63)0.189——Tail sign0.90 (0.47–1.70)0.738——Bone invasion1.55 (0.87–2.77)0.141——Nodal metastasis0.71 (0.18–2.79)0.623——Vascular/nerve invasion1.62 (0.86–3.06)0.136——Abbreviations: HR, hazard ratio; CI, confidence interval; LRFS, local recurrence–free survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001. Primary multivariable Cox model: N = 154, events = 50, C-index = 0.653. Forced covariates as in Table 2. MR features entered if univariable p < .10. Per the pre-specified Statistical Analysis Plan, peritumoral edema and peritumoral enhancement — overlapping visual manifestations of the peritumoral microenvironment (Fisher exact p < .001; VIF 1.5) — were not entered together; peritumoral edema, the feature with the stronger univariable signal across all three endpoints, was pre-specified as the single peritumoral-domain covariate eligible for multivariable entry (peritumoral enhancement excluded from primary model; the sensitivity model substituting peritumoral enhancement is summarized in the Results text). Ridge penalty λ = 0.01 Univariable and multivariable Cox regression of MR features in predicting local recurrence free survival Abbreviations: HR, hazard ratio; CI, confidence interval; LRFS, local recurrence–free survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001. Primary multivariable Cox model: N = 154, events = 50, C-index = 0.653. Forced covariates as in Table 2. MR features entered if univariable p < .10. Per the pre-specified Statistical Analysis Plan, peritumoral edema and peritumoral enhancement — overlapping visual manifestations of the peritumoral microenvironment (Fisher exact p < .001; VIF 1.5) — were not entered together; peritumoral edema, the feature with the stronger univariable signal across all three endpoints, was pre-specified as the single peritumoral-domain covariate eligible for multivariable entry (peritumoral enhancement excluded from primary model; the sensitivity model substituting peritumoral enhancement is summarized in the Results text). Ridge penalty λ = 0.01 MR features and metastasis-free survival Univariable MFS predictors were heterogeneous enhancement (HR, 3.1; P = .001), intratumoral necrosis (HR, 2.3; P = .002), vascular–nerve invasion (HR, 1.9; P = .02), maximum diameter (HR per 10-mm increment, 1.02 [95% CI: 1.01, 1.04]; P = .004), and male sex (HR, 1.8; P = .03); FNCLCC grade was not significant. In the multivariable model (n = 145; 60 events; C-index, 0.70), heterogeneous enhancement was a trend-level independent predictor (HR, 2.0 [95% CI: 0.9, 4.5]; P = .08); no feature reached the prespecified significance threshold, consistent with a lower event-per-variable ratio (Table 4). Table 4Univariable and multivariable Cox regression of MR features in predicting metastasis free survivalVariableUnivariableHR (95% CI)P valueMultivariableHR (95% CI)P valueAge (per year)1.01 (0.99–1.02)0.2561.00 (0.99–1.02)0.949Male sex1.78 (1.08–2.96) 0.025* 1.67 (0.96–2.91)0.067Maximum diameter (per 10 mm) 1.03 (1.01–1.04) 0.004* 1.03 (1.01–1.05)0.083FNCLCC grade 30.86 (0.49–1.50)0.5870.85 (0.45–1.59)0.613Translocation-driven histology0.95 (0.57–1.58)0.840——T1 heterogeneous signal1.38 (0.84–2.29)0.205——T2 heterogeneous signal1.42 (0.80–2.52)0.230——Peritumoral edema1.35 (0.80–2.28)0.268—— Heterogeneous Gd enhancement 3.10 (1.56–6.16) 0.001* 2.02 (0.91–4.47) 0.082 Peritumoral enhancement1.00 (0.60–1.66)0.994——Intratumoral necrosis2.28 (1.37–3.79) 0.002* 1.41 (0.78–2.58)0.257Intratumoral hemorrhage1.27 (0.74–2.19)0.384——Ill-defined margin on T10.96 (0.41–2.26)0.934——Ill-defined margin on T20.97 (0.42–2.28)0.953——Tail sign1.37 (0.80–2.34)0.251——Bone invasion1.58 (0.95–2.63)0.0771.31 (0.75–2.30)0.348Nodal metastasis2.04 (0.88–4.74)0.0961.90 (0.75–4.82)0.174Vascular/nerve invasion1.94 (1.13–3.33) 0.016* 1.28 (0.68–2.39)0.446Abbreviations: HR, hazard ratio; CI, confidence interval; MFS, metastasis-free survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001. Primary multivariable Cox model: N = 145, events = 60, C-index = 0.695. Forced covariates as in Table 2. Ridge penalty λ = 0.01 Univariable and multivariable Cox regression of MR features in predicting metastasis free survival Abbreviations: HR, hazard ratio; CI, confidence interval; MFS, metastasis-free survival; Gd, gadolinium; FNCLCC, Fédération Nationale des Centres de Lutte Contre le Cancer. Significance markers: * p < .05, †p < .001. Primary multivariable Cox model: N = 145, events = 60, C-index = 0.695. Forced covariates as in Table 2. Ridge penalty λ = 0.01 Histology-adjusted sensitivity analysis When translocation-driven histology was added as a fifth covariate, heterogeneous enhancement remained an independent OS predictor (HR, 3.0 [95% CI: 1.2, 7.9]; P = .02); peritumoral edema attenuated to a trend for OS (HR, 1.8; P = .10) after exclusion of six patients with uncertain histologic class but retained independence for LRFS (HR, 2.4 [95% CI: 1.2, 4.9]; P = .01). Translocation-driven histology itself was not independently prognostic for any endpoint. A triple-forced model (Additional Table A2) simultaneously entering age, sex, maximum diameter, FNCLCC grade 3, heterogeneous enhancement, surgical margin, and translocation-driven histology confirmed heterogeneous enhancement as an independent predictor of OS and MFS and a trend-level predictor of LRFS. Histology grade-surrogate analysis Logistic regression of FNCLCC grade 3 on the 13 MR features, adjusted for age, sex, and maximum diameter, showed poor grade discrimination (AUC, 0.59); no MR feature was independently significant (all P > .10), including heterogeneous enhancement (odds ratio, 0.74; P = .22), and peritumoral edema (odds ratio, 1.18; P = .68). The same panel modestly discriminated translocation-driven from complex-karyotype histology (AUC, 0.71), driven predominantly by younger age in translocation-driven cases (Additional Table A3). Discussion In 158 adults with contrast-enhanced MRI of newly diagnosed soft-tissue sarcoma, two binary visual descriptors — peritumoral edema and heterogeneous gadolinium enhancement — were independently and complementarily associated with shorter overall and local recurrence-free survival. Both findings held after adjustment for age, sex, tumor size, FNCLCC grade, and, in sensitivity analysis, translocation-driven histology. Neither FNCLCC grade nor translocation-driven class was independently prognostic, and a 13-feature MR panel failed to discriminate grade (AUC, 0.59), arguing against a latent grade-surrogate interpretation for either descriptor. The OS hazard ratio of heterogeneous enhancement is comparable in magnitude to the deep-radiomic risk clusters reported in recent series [23, 24], and the qualitative descriptor is independent of the translocation-driven versus complex-karyotype distinction [11, 25]. These results identify two compact, readily reported MR phenotypes — a peritumoral signal indexing local biologic aggressiveness and an intratumoral vascular-heterogeneity signal indexing systemic aggressiveness — that together outperformed conventional histologic-grade and molecular stratifiers in a translocation-driven enriched cohort. Three aspects of these findings extend prior work rather than restate it: which MR descriptors retained independent value depended on case-mix in this translocation-enriched cohort; the two descriptors showed an endpoint-specific division of labor, with peritumoral edema predicting local recurrence and heterogeneous enhancement predicting overall and metastatic outcome; and neither acted as a surrogate for histologic grade, since a 13-feature panel did not discriminate grade (AUC, 0.59). These descriptors are pan-histologic imaging traits rather than markers specific to translocation-driven tumors; their value here lies in retaining prognostic information in a population in which histologic grade and molecular class did not. Heterogeneous gadolinium enhancement has a direct pathophysiologic rationale: regions of perfusion mismatch on post-contrast imaging reflect disorganized angiogenesis, variable interstitial pressure, and microscopic necrosis — hallmarks of biologic aggressiveness. This interpretation aligns with the dynamic contrast-enhanced (DCE) MRI literature: Li et al. demonstrated that DCE pharmacokinetic parameters, particularly K-trans and Ve, track with histologic grade and short-term outcome in soft-tissue sarcoma [26]. Our result extends these findings by showing that a simple binary visual call, recorded on routine post-gadolinium images without pharmacokinetic modeling, captures much of the prognostic signal that DCE parameters describe quantitatively. The failure of a 13-feature MR panel to discriminate histologic grade (AUC, 0.59) further supports that heterogeneous gadolinium enhancement is not acting as a grade surrogate; its prognostic signal likely reflects a vascular-disorganization trait that crosses histologic and molecular lines. Pragmatically, this means the pattern can be incorporated into standard MR reports without specialized acquisitions, extending access beyond centers with DCE expertise (Fig. 2). The endpoint-specific profile is clinically informative: heterogeneous gadolinium enhancement reflects a systemic-risk phenotype, and peritumoral edema reflects a local biologic-aggressiveness phenotype. The dominant role of surgical factors in local control is well established [27]. Combining the two descriptors stratifies patients into four pragmatic risk groups; the highest-risk group (both features positive) carried the lowest 5-year OS and LRFS estimates, whereas the lowest-risk group (both features negative) carried the highest. The independent prognostic value of peritumoral edema is consistent with the edema-radiomics literature [28, 29], which has argued that the peritumoral region — not just the tumor core — contains biologically active tissue whose imaging features shape outcome. Peritumoral T2 hyperintensity in soft-tissue sarcoma likely reflects a composite of microscopic tumor infiltration beyond the gross margin, lymphatic obstruction, and neovascular leakage, all of which plausibly drive local recurrence [30]. The joint four-group stratification in Fig. 2 makes this division of labor visible: heterogeneous enhancement alone conferred the second-lowest 5-year overall survival (64%) but the second-highest local recurrence–free survival (75%), whereas peritumoral edema alone showed the reverse (81% and 58%). This discordant ranking is most parsimoniously explained by two biologically distinct processes — peritumoral edema indexing a peripheral, infiltrative phenotype that drives local relapse, and heterogeneous gadolinium enhancement indexing intratumoral vascular disorganization that drives distant progression and death. Operationally, peritumoral edema flags risk of local recurrence and supports heightened local surveillance, whereas heterogeneous gadolinium enhancement flags systemic risk and supports intensified metastatic screening. Inter-reader agreement for both descriptors was substantial (κ = 0.86), supporting reproducibility in routine practice on conventional sequences without specialized postprocessing (Fig. 3). A 48-year-old woman with synovial sarcoma of the anterior chest wall. (A) Axial T2-weighted image without fat suppression, (B) pre-contrast T1-weighted image, (C) post-gadolinium T1-weighted image. The tumor demonstrates homogeneous enhancement. No peritumoral edema is present. The patient was alive at 8-year follow-up. A 58-year-old woman with undifferentiated pleomorphic sarcoma of the posterior chest wall. (D) Axial T2-weighted image with fat suppression, (E) pre-contrast T1-weighted image, (F) post-gadolinium T1-weighted image. The tumor demonstrates heterogeneous enhancement with medial surrounding muscle edema (peritumoral edema) on T2-weighted imaging. The patient died of disease at 14 months Several MR-prognostic signals reported in earlier cohorts attenuated in our analysis. Crombé et al. [9] reported that peritumoral enhancement, intratumoral necrosis, and more than 50% T2 heterogeneity were independently associated with metastasis; in our translocation-driven enriched cohort, these descriptors attenuated after adjustment, while heterogeneous gadolinium enhancement retained significance. Peeken et al. and Yan et al. [31, 32], reported MR-radiomic grade classifiers with validation AUCs of 0.78–0.88; our grade logistic AUC of only 0.59 demonstrates that those pipelines operate in a different part of the feature space than our case-mix permits. Fernebro et al. [9] reported that diffusely infiltrative peripheral growth on MRI was strongly associated with recurrence in a population-based Scandinavian cohort dominated by complex-karyotype histologies, yet tumor margin was not prognostic in our translocation-driven enriched cohort. Instead, peritumoral edema and enhancement better reflected tumor behavior and predicted patient outcomes than did histology or tumor margin on T1- or T2-weighted sequences [33, 34]. For example, alveolar soft-part sarcoma characteristically displays infiltrative margins that reflect its distinctive vascular biology rather than conventional prognostic determinants [35]. The most parsimonious explanation is case-mix: translocation-driven tumors are younger, smaller, and less frequently necrotic, and the descriptors that survive multivariable adjustment depend on the histologic substrate. Three features of our cohort help explain the divergence: the mean age (48 years) is roughly a decade younger than typical adult STS series; 40% of patients had translocation-driven tumors — roughly double the proportion in many Western tertiary referral cohorts; and restriction to contrast-enhanced examinations enriched for larger tumors [11]. Together, these features explain the preserved signal for heterogeneous gadolinium enhancement (a pan-histologic vascular-disorganization trait) and the attenuation of descriptors that track more tightly with complex-karyotype biology. Clinical implications Our findings support adding two binary descriptors — presence of peritumoral edema and presence of heterogeneous gadolinium enhancement — to standard preoperative MR reports in soft-tissue sarcoma. Both can be recorded rapidly on conventional sequences, require no specialized acquisitions, and demonstrated substantial inter-reader agreement. Operationally, peritumoral edema supports intensified local surveillance (closer-interval MRI; lower threshold for biopsy of suspicious peritumoral changes), and heterogeneous gadolinium enhancement supports closer metastatic screening. Combining both descriptors stratifies patients into four pragmatic risk groups that could inform individualized surveillance intensity, particularly in translocation-driven tumors where FNCLCC grade offers limited prognostic resolution. Prospective multicenter validation is required before these descriptors inform treatment-escalation decisions. Limitations This study has limitations. It was retrospective, single-center, and modest in size. Our institution is a tertiary sarcoma referral center; patients reached the service predominantly through external surgical referral rather than a population-based catchment, enriching the cohort for young adults (mean age, 48 years), high-grade disease (79% grade 3), and translocation-driven histology (40%) relative to population-based series, in which FNCLCC grade 3 typically accounts for 30–50% and translocation-driven tumors for roughly 20%. This referral-driven case-mix both defines the strength of our argument and limits direct generalizability to unselected cohorts; findings should be confirmed in population-based or multicenter series. Because the study was designed and powered for the full enriched cohort, it was not powered for a translocation-restricted analysis; subgroup-specific effect estimates within the translocation-driven tumors await a larger multicenter cohort. MR acquisitions spanned two field strengths and a 16-year period; although this reflects real-world practice, unmeasured protocol differences may have introduced noise. Restriction to contrast-enhanced examinations excluded 22 patients and may have further enriched for larger tumors. Confidence intervals for several effect estimates were wide; prospective multicenter validation is needed. Finally, we did not include quantitative DCE or diffusion-weighted parameters; integrating those with the two visual descriptors may further improve risk stratification. Conclusion In 158 adults with contrast-enhanced MRI of newly diagnosed, translocation-driven enriched soft-tissue sarcoma, peritumoral edema and heterogeneous gadolinium enhancement were independent, complementary MR predictors of outcome. Peritumoral edema indexed local biologic aggressiveness; heterogeneous gadolinium enhancement indexed systemic risk. Both were independently prognostic where FNCLCC grade and translocation-driven histology were not in this cohort, and neither acted as a grade surrogate. Incorporating these two binary descriptors into standard MR reports could inform risk-stratified surveillance in this subgroup. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Not applicable. Abbreviations Area under the receiver operating characteristic curve Confidence interval European Society for Medical Oncology Fédération Nationale des Centres de Lutte Contre le Cancer Gadolinium Heterogeneous gadolinium enhancement Hazard ratio Interquartile range Local recurrence–free survival Metastasis-free survival Magnetic resonance imaging Overall survival Peritumoral edema Standard deviation Soft-tissue sarcoma Author contributions C.J.L. and H.Y.L. conceived and designed the study. H.Y.L. and C.J.L. drafted the manuscript. H.Y.L., C.L.W., C.A.L., H.T.W., P.T.L., and C.J.L. performed the MRI feature scoring and image analysis. C.H.C. provided histopathologic diagnosis, molecular classification, and pathology-related interpretation. C.C.Y. contributed clinical oncology data, treatment information, and follow-up outcomes. C.M.C. and P.K.W. contributed surgical data, resection margin status, and local recurrence outcomes. P.T.L.,H.Y.L. and C.J.L. performed the statistical analysis and data interpretation. C.L.W., C.A.L., and H.T.W. critically reviewed the imaging methodology. C.H.C., C.C.Y., C.M.C., and P.K.W. critically revised the manuscript for important intellectual content. C.J.L. supervised the project and is responsible for the integrity of the work as a whole. All authors reviewed and approved the final version of the manuscript. Funding This project was supported by the National Science and Technology Council (NSTC 114-2314-B-A49-996) and Taipei Veterans General Hospital (V114D78-003-MY2-2, V114D78-002-MY2-2, and V114D78-001-MY2-2). Data availability All data generated or analysed during this study are included in this published article and its supplementary information files. Declarations Ethics approval and consent to participate This retrospective single-institution study was approved by Taipei Veterans General Hospital review board (IRB Number:2024-09-010AC), which waived the requirement for informed consent. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data generated or analysed during this study are included in this published article and its supplementary information files.