Title MRI based risk stratification of uterine2025 Zlotykamien-Taieb E, Gherman D, Al Rouhban R, Thomass.pdf ✅

included with the group of malignancy and sarcoma as in previous studies [15]. Malignant sarcomas are also a very heterogeneous group and they are divided into four pathologic categories: (a) homologous composed only of uterine tissue; (b) heterologous types with mixed uterine and non-uterine tissues, usually striated muscle, bone, and cartilage; (c) pure which include only mesodermal structures; and (d) mixed types with mesodermal and non-mesodermal structures; mainly epidermal [13]. 2.5. Statistical analysis Statistical analyses were conducted using MedCalc software (version 20.0, MedCalc Software Ltd., Belgium). Descriptive analysis was per- formed using a non-parametric Mann–Whitney test for continuous var- iables; Fisher’s exact test was used for categorical or nominal variables based on the most experienced reader’s analysis (reader 2). Odds ratios (OR) for predicting malignancy were calculated with 95 % confidence intervals and p-values for each predictor variable for malignancy. A multivariate regression model was built on the basis of criteria with the higher univariate odd ratios to propose a modified algorithm. Quadratic κ coefficients were used to assess intra- and inter-observer agreements for lesion characterization, with a kappa value indicating agreement levels from poor (<0.2), fair (0.2–0,4), moderate (0.4–0.6), substantial (0.6–0.8) to almost perfect (>0.8). Univariate analysis was used to calculate odds ratios for each predictor variable for malignancy. Receiver operating characteristic (ROC) curve analysis compared the results of original and modified algorithms, and the McNemar test compared the accuracy between algorithms. A p-value of < 0.05 was considered statistically significant. 3. Results 3.1. Population and methods The study included 2665 women with mesenchymal tumors who underwent pathological analysis at two centers during the period screened. Exclusion criteria encompassed patients without available MRI examination or with incomplete acquisition protocol (n = 2037), MRI examinations performed more than 6 months prior to surgery or percutaneous biopsy (n = 142), and cases where surgery was conducted for extra-myometrial pathologies with incidental findings of myometrial disorders (n = 31). The final cohort comprised 455 women with mesenchymal tumors, with a mean age of 43 years (range: 15–82 years) (Fig. 1), 62 of these women were menopausal (13,2%) (Table S1). Final pathology was confirmed post-surgery in 452 cases and after percuta- neous biopsy followed by 6 months of imaging follow-up in 3 cases. The diagnoses were established by pathology, including hysterectomy (59.8 %; 272/455), laparotomic myomectomy (25.3 %; 115/455), hystero- scopic myomectomy (7.7 %; 35/455), laparoscopic myomectomy (5.3 %; 24/455), or surgical or percutaneous biopsy (2 %; 9/455). To ensure the correlation between the target lesion and pathology after partial surgery such as myomectomy, we compared MR size and pathological size. The median delay between MR imaging and surgery or percuta- neous biopsy was 51.5 days (range: 1–181 days). Gadolinium-enhanced sequences were available for 94/437 leiomyomas and 17/18 STUMP or malignant UMT. Histopathological analysis revealed that 96 % of cases were leiomyomas (437/455), 0.4 % were STUMP (2/455), and 3.5 % were malignant UMT (16/455) (Table 1). 3.2. Patient demographics Patient age was significantly higher in individuals with STUMP or Fig. 1. Flowchart showing patient population and index test results. Table 1 Histopathologic parameters. Histopathological subtypes N = 455 Benign leiomyoma (n = 437) 96 % (437) Usual leiomyoma 71 % (323) Leiomyoma with degeneration 22.7 % (103) Cellular leiomyoma 1.3 % (6) Leiomyoma with bizarre nuclei 0.4 % (2) Lipoleiomyoma 0.2 % (1) Vascular leiomyoma 0.2 % (1) Leiomyoma + intravascular disseminated leiomyomatosis 0.2 % (1) Tumor of uncertain malignant potential and leiomyosarcoma 3.9 % (18) Leiomyosarcoma 3.3 % (15) Undifferentiated uterine sarcoma 0.2 % (1) Smooth muscle tumor of uncertain malignant potential (STUMP) 0.4 % (2) E. Zlotykamien-Taieb et al. European Journal of Radiology 187 (2025) 112126 3
malignant UMT (mean: 56 years, range: 33–82) compared to those with benign UMT (mean: 43 years, range: 15–74) (P < 0.001). Menopausal status was more prevalent in patients with STUMP or malignant UMT (66.7 %; 12/18) compared to those with benign UMT (11.4 %; 50/437) (P < 0.001; OR = 15.5, 95 % CI 5.6–43.0). No significant difference exists in gynecological symptoms between benign UMT and STUMP or malignant UMT (Tables S1 and S2). 3.3. Interobserver agreement Interobserver agreement was substantial or almost perfect for all criteria except for lymph node assessment (κ = 0.566), which displayed moderate agreement. The intraclass correlation coefficient for ADC value evaluation was high (ICC = 0.914) (Table 2). 3.4. Lesion characterization Univariate analysis of features included in the original model showed that abnormal lymph nodes were more frequently observed in STUMP or malignant UMT (22.2 %; 4/18) compared to benign UMT (1.4 %; 6/437) (P = 0.003; OR = 20.5, 95 % CI 5.2–80.9). No difference was noted regarding peritoneal implants, although this may be related to small sample size since these were observed in only two patients. Intermediate T2-weighted signal was observed in all STUMP or malignant UMT (100 %; 18/18) while about one-third of benign UMTs exhibited intermediate T2W signal (35.5 %; 155/437) (P < 0.001; OR = 67.2, 95 % CI 4.0–1123.1). Importantly, no STUMP or malignant UMT displayed a low T2W signal. High signal on diffusion (equal to or higher than the endometrium) was more frequently observed in STUMP or malignant UMT (77.8 %; 14/18) compared to benign UMT (15.5 %; 68/437) (P < 0.001; OR = 16.05, 95 % CI 5.13–50.3). High signal on diffusion (higher than the bladder) was present in all STUMP or malignant UMT (100 %; 18/18) and significantly more than in benign UMT (33.6 %; 147/437) (P < 0.001; OR = 72.87, 95 % CI 4.4–1217.7). ADC values were lower in STUMP or malignant UMT (mean = 0.76 × 10− 3 mm2 /sec, 95 % CI 0.67–0.90) than in benign UMT (mean = 1 × 10− 3 mm2 /sec, 95 % CI 0.98–1.07) (P < 0.001). No STUMP or malignant UMT displayed an ADC value > 1.23 × 10− 3 mm2 /sec (Fig. 2). 3.5. Other criteria The myometrial mesenchymal tumor was more frequently unique in STUMP or malignant UMT (61.1 %; 11/18) compared to benign UMT (18.3 %; 80/437) (P < 0.001; OR = 7.0, 95 % CI 2.6–18.7). Irregular tumor margins were significantly more prevalent in STUMP or malig- nant UMT (83.3 %; 15/18) than in benign UMT (4.6 %; 20/437) (P < 0.001; OR = 104.3, 95 % CI 27.9–389.6). Intra-tumoral hemorrhage was more common in STUMP or malignant UMT (55.6 %; 10/18) than in benign UMT (13.7 %; 60/437) (P < 0.001; OR = 7.9, 95 % CI 2.98–20.69). Tumor size, endometrial stripe thickness, pelvic fluid, peritoneal implants, and enlarged pelvic lymph nodes were not discriminative in assessing malignancy (Table 3). 3.6. External validation and model refinement The original model for distinguishing benign from malignant UMT including STUMPs was evaluated using criteria including enlarged pel- vic lymph nodes, T2W signal intensity, DW signal intensity compared to endometrium, and an ADC cutoff value of 0.9 × 10− 3 mm2 /sec. The model correctly classified benign and STUMP or malignant UMT in 420 out of 455 cases, resulting in an accuracy of 92.5 % (95 % CI 93,1–94,3) for detecting STUMP and malignant UMT, with a sensitivity of 61.1 % (95 % CI 38.5–83.6) and specificity of 93.8 % (95 % CI 91.2–95.8). Fig. 3 details the false positive and false negative cases at each step of the original algorithm. A modified new model was tested, based on a multivariate regression analysis model allowing to incorporate two additional criteria: irregular tumor margins and menopausal status, along with modifications to two criteria: DW signal compared to bladder and an ADC cutoff value of 1.23 × 10− 3 mm2 /sec. This refined new model achieved correct classification in 446 out of 455 cases, resulting in an accuracy of 98 % (95 % CI 97.1–98.1 %), with a sensitivity of 83.3 % (95 % CI 79–88) and speci- ficity of 98.6 % (95 % CI 98–99) (Table 4). Fig. 4 details the false positive and false negative cases at each step of the modified algorithm. The modified algorithm was significantly more accurate than the initial al- gorithm (P = 0.001), allowing to increase the number of correctly classified cases by 5.49 % (25/455) (Table 4). Based on this modified multivariate analysis model, we propose a 5- category scoring system (Fig. 4). 4. Discussion Our study demonstrates that combining 5 criteria based on multi- variate analysis in a modified algorithm (T2W signal, DW signal, ADC cut off value of 1.23 x 10–3 mm2/sec, tumor margins and menopausal status) allows to distinguish benign from malignant UMTs with an ac- curacy of 98 % (CI95% 97,1–98,1), a sensitivity of 83.3 % (CI95% 79–88) and a specificity of 98.6 % (CI95% 98–99). This model allows to build a stratification score that would help in the management of atypical uterine tumors. Leiomyoma, a common uterine mesenchymal tumor, affects more than 50 % of Caucasian women and up to 80 % of women of African descent, and is the leading cause of hysterectomy [16,17]; In contrast, uterine sarcomas are very rare, with an estimated incidence of three to seven per 100,000 women [18], accounting for only 8 % of all malignant tumors of the uterine corpus [19]. Leiomyosarcoma, representing 60 % of uterine sarcomas, is associated with a poor prognosis even when diagnosed at an early stage leading to a significant increase in uterine cancer-associated deaths [20–24]. The recurrence rate reaches 53–75 %, even in the initial stages of the disease, with locoregional or distant recurrences within the first two years after diagnosis [20,21]. Overall survival expectancy is 2.6 years, and 2-, 5-, and 10-year survival are 57 %, 24 %, and 12 %, respectively [20,21]. Our study’s incidence of leiomyosarcomas was consistent with previous research, highlighting the importance of preoperative risk assessment to guide surgical de- cisions, especially considering the risks associated with morcellation in case of preoperative underdiagnosis. In the current study, the prevalence of malignant UMT aligns with previous series [19]) with an incidence of 0.35 % in premenopausal and 0.57 % in postmenopausal women [25]. Gynecologic surgeons must consider the preoperative risk of sarcomas to decide between radical and conservative surgery, especially regarding infertility and the risks associated with morcellation, even when performed in an endoscopic bag. Morcellation, a common method for removing leiomyomas [26], significantly impacts survival and recurrence rates if an occult Table 2 Inter observer agreement between senior and junior readers. ICC Kappa values (± SD) Abnormal lymph nodes 0.566 ± 0.12 Intermediate or high T2W SI 0.899 ± 0.49* Heterogeneous T2W SI 0.899 ± 0.50† Irregular margins 0.724 ± 0.28 High DW signal 0.88 ± 0.50† High DW SI >= endometrium 0.807 ± 0.40 ADC value 0.914 Hemorrhagic changes 0.89 ± 0.36 Heterogeneous enhancement 0.937 ± 0.48 Areas of non-enhancement 0.74 ± 0.4 T2W: T2 weighted signal imaging, DW: diffusion weighted, SI: signal intensity. * Numbers are linear Kappa values (+/-SD). † Numbers are quadratic Kappa values (+/-SD). E. Zlotykamien-Taieb et al. European Journal of Radiology 187 (2025) 112126 4