This case-control study was approved by the Ethical Committee of Children's Hospital of Shanghai/Shanghai Children's Hospital, Shanghai Jiao Tong University (Approval number: 2022R160-E01), and informed consent was waived due to the retrospective nature of the analyses and the use of anonymized medical records.

From January 2011 to December 2022, 89 individuals with soft tissue RT or RMS were identified after a search of the medical files and the picture archiving and communication systems (PACS) at our institution (Figure 1). The inclusion criteria for this study were as follows: (a) confirmation of soft tissue RT or RMS by surgical pathology; (b) multidetector CT and/or MRI examination was performed within 6 months before surgery and chemotherapy. Exclusion criteria were as follows: (a) lesions located in the internal organs or central nervous system; (b) patients did not undergo CT or MRI examination before locoregional treatment; (c) image quality did not meet the diagnostic needs. Ultimately, 49 patients with soft tissue tumors, comprising 33 patients with RMS and 16 patients with soft tissue RT, were enrolled in this study. Data were collected on patient demographics, clinical symptoms, and immunohistochemistry.

Among the 49 patients, 45 underwent multi-slice spiral CT scans, 19 underwent MRI scans, and 15 underwent both CT and MRI scans. All of the patients fasted for four hours to before the scan, and those who couldn't cooperate with the CT or MRI tests were given 10% chloral hydrate (0.5 ml/kg) orally for sedation prior to the test.

The CT examinations were performed using 64-multidetector CT scans (Lightspeed Ultra 64, GE Medical Systems, USA or Aquilion Ultra 64, TOSHIB, Japan), with the following parameters: a layer thickness of 0.625 mm, interval of 2.5 mm, pitch of 0.984, tube voltage of 100 kV, and tube current of 240 mA. Coronal and transverse multiplanar reconstruction CT images were reconstructed with 3 mm section thickness. After the acquisition of unenhanced CT, 2 ml/kg of Iohexol (350 mg/ml, Schering, Germany) was administered intravenously at a rate of 1.8–2.5 ml/s using an automatic power injector (OptiVantage DH; Mallinckrodt, St. Louis, Mo) into a 22- or 24-gauge intravenous cannula, without normal saline bolus administration. Trip-phase enhanced images were obtained at 15–20 s, 40–50 s, and 120–180 s after intravenous contrast injection.

All patients used a 3.0 T scanner (Ingenia, Philips Medical Systems, Best, the Netherlands). To achieve optimal resolution, coils were selected according to the respective body region examined. Scanning methods: conventional axial, coronal, and sagittal scans, including T1-weighted images (T1WI), T2-weighted images (T2WI) and fat-suppressed T2-weighted images (Fs-T2WI), with a slice thickness of 5 mm and an interval of 4 mm. A single-shot echo sequence was used to generate diffusion-weighted images (DWIs). The diffusion gradients were applied in three orthogonal directions (x, y, and z). After 0.2 ml/kg Gd-DTPA (3 ml/s) was injected intravenously, sagittal, axial, and coronal T1WI pictures were acquired. Detailed scan parameters are shown in Table 1.

The CT and MRI pictures were assessed by two radiologists with more than ten years of expertise who were blinded to the ultimate diagnosis. Individual interpretation differences were settled through consensus. The following imaging characteristics were noted and documented: (a) number (single or multiple); (b) location (head and neck/trunk/extremities); (c) size (maximum tumor diameter); (d) morphology (multilobulated/round/others); (e) margin (clear or unclear); (f) density/signal intensity (lower or higher than normal muscle/cystic necrosis/calcification/hemorrhage/fat); (g) peripheral cyst (present or absent); (h) surrounding blood vessels (present or absent); (i) rim enhancement (present or absent). Other radiologic features, such as adjacent bone destruction and metastatic sites, were also evaluated. The largest tumor diameter was measured in any axial, coronal, or sagittal plane on both CT and MRI. An unclear margin was defined as lacking clear borders between the tumor and surrounding structures on unenhanced images.

DWI examinations were performed on three cases of soft tissue RT and thirteen cases of RMS. The ADC maps were automatically created for the obtained images with the software built into the MR unit using b values of 0 and 1,000 s/mm². All MRI data were then transferred to an offline workstation in DCM format. Two experienced pediatric radiologists, with 10 and 20 years of MRI experience, independently assessed the MRI images and identified the regions of interest (ROI), which included the entire tumor, for histogram analysis using MaZda software (version 4.6, The Technical University of Lodz, Institute of Electronics, http://www.eletel.p.lodz.pl/mazda/). The software automatically extracted nine histogram parameters, such as the mean, variance, skewness, and kurtosis, as well as the 1st (ADCp1), 10th (ADCp10), 50th (ADCp50), 90th (ADCp90), and 99th (ADCp99) percentiles of ADC.

Statistical analyses were conducted using SPSS Statistics version 25 (IBM) and R version 4.2.2 (http://www.r-project.org) statistical software. The Shapiro-Wilk test was used to determine data normality, and Levene's test was used for homogeneity of variance. The mean and standard deviation (SD) were used to describe continuous variables with a normal distribution, and compared with the student t-test. The median and interquartile ranges (IQRs) were used to describe continuous variables with a non-normal distribution, and compared with the Mann-Whitney U-test. The chi-square or Fisher's exact test was used to compare categorical variables, and the results are expressed as numbers and percentages of patients. The variables of imaging with p < 0.05 in univariate analysis were included in the multivariate logistic regression analysis (forward LR) to determine the independent factors for distinguishing soft tissue RT and RMS. For each risk factor, odds ratios (OR) were calculated as relative risk estimates with 95% confidence intervals (CI). Additionally, Cohen's kappa was used to calculate the inter-observer agreement between the two radiologists. A two-tailed p < 0.05 was considered to be statistically significant.

Combining independent predictors identified by multivariate logistic regression (forward LR) to form a model for discriminating between soft tissue RT and RMS. To verify the accuracy of the model, bootstrapping resampling was performed 1,000 times, and the results were visualized using nomograms. The sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate the model's discrimination. Calibration curve analysis was used to assess the consistency between the tumor types identified by the model and the actual tumor types. Additionally, decision curve analysis (DCA) was conducted to determine the clinical utility of the model by assessing the net benefits at various threshold probabilities. Kaplan–Meier (K–M) survival curves with the log-rank test were generated to evaluate the differences in survival rate between the two groups.

The clinical and immunohistochemical findings of soft tissue RT (n = 16) and RMS (n = 33) are presented in Table 2. There was no statistical difference in age (p = 0.518) or sex (p = 0.08) between the two groups. The incidence of limited motion of extremities in the soft tissue RT group was significantly higher than the RMS group (25% vs. 0%, p = 0.015). Immunohistochemically, the CK and EMA expression rates were significantly higher in soft tissue RT compared to RMS (93.3% vs. 13.8%, 92.9% vs. 15.0%, respectively, all p < 0.001). Of note, INI was expressed by all patients with RMS but not by those with soft tissue RT (100% vs. 0%, p < 0.001). Also, immunohistochemistry markers like MyoG, MyoD1, and DES were more often found to be positive in RMS patients than in soft tissue RT patients (all p < 0.001). The Ki-67 levels of soft tissue RT and RMS were 45.1 ± 28.7 and 57.8 ± 20.9, respectively (p = 0.095).

There were no statistically significant differences in terms of tumor number, location, size, or shape between the two groups. The peripheral cyst (50% vs. 24.2%, p = 0.071), hemorrhage (12.5% vs. 12.1%, p > 0.999), and fatty degeneration (0% vs. 6.1%, p > 0.999) also did not differ between the soft tissue RT and RMS. Unclear margin and calcifications were more commonly found in the soft tissue RT group than in the RMS group (87.5% vs. 57.6%, 43.8% vs. 12.1%, respectively, all p < 0.05). Surrounding blood vessels and rim enhancement were more frequently observed in patients with RMS compared with soft tissue RT (51.5% vs. 18.8%, 63.6% vs. 18.8%, respectively, all p < 0.05). As shown in Table 3.

On unenhanced CT images, thirteen patients with soft tissue RT exhibited iso- or hypodense relative to adjacent muscles (Figure 2), and only one exhibited hyperdense, which was not significantly different compared with RMS (Figure 3) (all p > 0.05). According to MRI characteristics, most cases in the two groups demonstrated Iso- or hypointense signal on T1-weighted MR images, and hyperintense signal on T2-weighted MR images (Figures 4, 5). Furthermore, none of ADC histogram parameters were statistically significant (all p > 0.05). Two radiologists evaluated subjective image indicators with an excellent degree of agreement (kappa values: 0.81–1).

In univariate analysis, four imaging features were significantly different between the two groups at a test level of p < 0.05 (Table 4). Based on multivariate logistic regression analysis (forward LR), an unclear margin (p-value, adjusted odds ratio [95% confidence interval]: 0.03, 7.96 [1.23, 51.67]), calcification (0.012, 30.37 [2.09, 440.70]) were independent differential factors for predicting soft tissue RT over RMS; rim enhancement (0.007, 0.05 [0.01, 0.43]) were independent differential factors for predicting RMS over soft tissue RT.

The nomogram was developed using the significant variables to make predictions about the occurrence of soft tissue RTs (Figure 6A). The results of further ROC curve analysis revealed that the area under the curve (AUC) was 0.872 with 81.8% specificity and 81.3% sensitivity, indicating excellent diagnostic performance (Figure 6B). Calibration curves (Figure 6C) demonstrated that the nomogram accurately estimated the agreement between observed outcome frequencies and predicted probabilities. Additionally, the DCA curve (Figure 6D) showed a high clinical net benefit over the “treat-all” or “treat-none” strategies and was of clinical application.

The final follow-up date was December 2022, with a median follow-up time of 53 months. Clinical assessment and follow-up imaging revealed that recurrences occurred in 6.3% (1/16) of soft tissue RT patients and 21.2% (7/33) of RMS patients, with a mean time to recurrence of 19 vs. 16 months after diagnosis. Among the 43.8% (7/16) patients in the soft tissue RT group who had metastasis, 57.1% (4/7) had lung metastasis and 42.9% (3/7) had other metastatic locations, such as bone, lymph nodes, the omentum, etc. However, the bone (3/10) is the most common metastatic site for RMS patients, followed by the lung, lymph nodes, central nervous system, etc. The 3-year OS rate of the patients with RT were significantly lower than those of the patients with RMS (22.6% vs. 90.8%, p < 0.01). Furthermore, the median survival time for RT patients is only 14 months (Figure 7).