Quantitative Time-Dependent Diffusion MRI for Diagnosis and Aggressiveness Assessment of Endometrial Cancer: A Prospective Study

Preoperative differentiation of benign and malignant endometrial lesions, along with the identification of aggressive histological types of endometrial cancer (EC), is crucial for guiding treatment strategies. Time-dependent diffusion magnetic resonance imaging (TDD-MRI), which allows the characterization of tissue microstructure at the cellular level, is not currently applied for endometrial lesions. This study aimed to evaluate TDD-MRI-derived microstructural parameters for noninvasively distinguishing benign and malignant endometrial lesions and predicting aggressive histological types of EC.

This prospective study enrolled 177 patients with clinically suspected EC who underwent TDD-MRI between January 2024 and March 2025. The Imaging Microstructural Parameters Using Limited Spectrally Edited Diffusion

was used to extract microstructural parameters, including the cell diameter (d), intracellular volume fraction (v in), cellularity (number of cells per unit area), cellularity index (vin/d), and extracellular diffusivity (Dex), along with three apparent diffusion coefficient measurements. The area under the receiver operating characteristic curve (AUC) was used to assess diagnostic performance. The Pearson correlation coefficient between the microstructural parameters and histopathological measurements was calculated.

A total of 130 women (mean ± standard deviation age: 56 ± 14 years) administered uterine curettage or surgery were included in the final analysis. All microstructural parameters showed significant differences between benign endometrial lesions and EC (P < 0.05), as well as between nonaggressive and aggressive EC (P < 0.05). Cellularity exhibited the highest AUC of 0.86 for distinguishing benign endometrial lesions from EC, whereas the cellularity index showed the highest AUC of 0.88 for distinguishing aggressive histological types. D 0Hz was positively correlated with Dex (P < 0.05) and negatively correlated with diameter ( P < 0.05), cellularity index ( P < 0.01) and vin (P < 0.001) in patients with benign endometrial lesions. D0Hz was positively correlated with Dex (P < 0.001) and negatively correlated with vin (P < 0.001) in patients with EC. Microstructural parameters strongly correlated with corresponding pathological features ( r = 0.77–0.83; P < 0.001). Korean J Radiol 2025;26(10):973-985 eISSN 2005-8330 https://doi.org/10.3348/kjr.2025.0633 Original Article | Oncologic Imaging Received: May 19, 2025 Revised: July 1, 2025 Accepted: July 31, 2025 Corresponding author: Qi Yang, MD, PhD, Department of Radiology, Beijing Chaoyang Hospital, Capital Medical University, No. 8 Gongti South Road, Chaoyang District, Beijing 100020, China • E-mail: [email protected] Corresponding author: Hua Li, MD, PhD, Department of Gynecology and Obstetrics, Beijing Chaoyang Hospital, Capital Medical University, No. 8 Gongti South Road, Chaoyang District, Beijing 100020, China • E-mail: [email protected] This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https:// creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited. 974 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org and biophysical models to characterize tissues at the cellular level. Notably, Imaging Microstructural Parameters Using Limited Spectrally Edited Diffusion (IMPULSED) is clinically feasible, enabling cellular microstructure mapping in just 5–7 minutes [11]. Such novel microstructural information has been clinically applied for prostate [12], breast [11,13,14], brain [15], and ovarian [16] cancers. Previous evidence has shown the capability of TDD-MRI in differentiating benign and malignant tumors as well as pathological grading [12,13,15-17]. However, it remains unclear whether TDD-MRI can characterize endometrial tissues, noninvasively differentiate benign and malignant lesions, and classify aggressive histological subtypes of EC. Therefore, this study aimed to evaluate the diagnostic feasibility of TDD-MRI in endometrial disease, comparing it with conventional DWI measurements for differentiating benign and malignant endometrial lesions and identifying aggressive histological types of EC.

Study Participants Following the Declaration of Helsinki, this prospective study was approved by the Institutional Review Board of Beijing Chaoyang hospital of Capital Medical University (IRB No. 2023-11-13-1) and registered with the Chinese Clinical Trial Registry (ChiCTR2500095674). Totally, 177 patients with clinically suspected EC were enrolled to undergo MRI between January 2024 and March 2025. The inclusion criteria were as follows: 1) scheduled conventional contrast-enhanced pelvic MRI and TDD-MRI, and 2) willingness and ability to undergo MRI and provide informed consent. As the enrolled patients progressed through the study procedure, some were excluded from the final analysis based on the following criteria: 1) lack of pathological confirmation, 2) prior treatment for endometrial disease before MRI, 3) Insufficient MRI quality, and 4) final histology showing atypical endometrial hyperplasia. The participant flowchart is shown in Figure 1.

Endometrial cancer (EC) is a common gynecological malignancy with an increasing incidence and a younger demographic [1]. Abnormal uterine bleeding (AUB) is a common symptom of EC. However, benign endometrial lesions (e.g., endometrial polyps and hyperplasia without atypia) may also cause AUB [2]. Benign endometrial lesions are often managed with diagnostic curettage or conservative treatment, whereas EC typically requires hysterectomy. The International Federation of Gynecology and Obstetrics (FIGO) staging of EC identifies aggressive histological types [3]. Therefore, preoperative differentiation of benign and malignant endometrial lesions, along with the identification of aggressive EC histological types, is crucial for guiding subsequent patient treatment. Evidence suggests significant differences between the microenvironments of benign endometrium and EC [4,5]. In EC, the number of epithelial and endometrial stromal cells increases and decreases, respectively. Additionally, changes in immune cell populations are observed, with a reduced proportion of cytotoxic and naïve CD8 lymphocytes, and an increased proportion of CD4+ T regulatory cells in EC [6]. This highlights the significant cellular differences between endometrial diseases, emphasizing the value of cell-level differentiation for diagnosis. Currently, no noninvasive technique can detect these diseases at the cellular level before pathological confirmation. Diffusion-weighted imaging (DWI) is a noninvasive imaging technique that offers functional insights and enhances the morphological details provided by conventional magnetic resonance imaging (MRI) [6-8]. DWI allows the quantification of diffusion through the apparent diffusion coefficient (ADC), which reflects the physiological characteristics of tissue microcirculation [9]. However, DWI cannot provide microstructural parameters, such as intra- and extracellular space, cell size, and permeability [10]. Recent advancements in time-dependent diffusion (TDD)-MRI have demonstrated its unique ability to depict cellular microstructures. This diffusion MRI-based technique leverages multi-diffusion times, multi-b values,

TDD-MRI-derived microstructural parameters demonstrated high performance in differentiating benign from malignant endometrial diseases and identifying aggressive types of EC.

Magnetic resonance imaging; Time-dependent diffusion MRI; Endometrial cancer; Histological types; Microstructural parameters 975 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org Image Acquisition The IMPULSED strategy was used for diffusion MRI by applying oscillating gradient spin-echo (OGSE) and pulsed gradient spin-echo (PGSE) sequences [11]. Scanning was performed using a Philips 3T Elition MR scanner (Philips Healthcare, Best, the Netherlands) with an external pelvic phased-array coil. OGSE data were acquired at oscillating frequencies of 33 Hz (effective diffusion time, 7.5 ms; 2 cycles; b = 0, 80, 160, 250 s/mm 2) and 17 Hz (effective diffusion time, 15.0 ms; 1 cycle; b = 0, 200, 400, 600, 800 s/mm 2). PGSE data were acquired with a diffusion duration/separation = 60/82.3 ms at b-value of 500/1,000/1,500 s/mm 2. The PGSE sequence used represents conventional DWI, with the diffusivity value derived from this acquisition (at 0 Hz) corresponding to the standard ADC commonly used in clinical diagnosis [12,14]. Both sequences used the following parameters: repetition time/echo time, 3,000/143 ms; field of view, 220 mm x 220 mm; voxel reconstruction size, 2.80 x 2.80 x 6.00 mm 3; single-shot echo planar imaging; half scan factor, 0.684; water-fat shift (pixels)/bandwidth (Hz), 14.610 pixel/29.7 Hz; fat suppression with spectral adiabatic inversion recovery. Dynamic stabilization was used to minimize the DWI signal drifts. The scanning time for TDD-MRI was approximately 9 minutes 8 seconds and its protocol is summarized in Figure 2. Image Analysis The image analysis method used has been described previously [18]. Briefly, diffusion signals are modeled as arising from two distinct compartments so that S = v in x Sin + (1 - vin ) x Sex, where vin denotes the water volume fraction of the intracellular space, and Sin and Sex are intracellular and extracellular diffusion MRI signals, respectively. Cancer cells were modeled as impermeable spheres so that Sin can be described using analytical expressions and the cell diameter (d) can be estimated to represent the mean cell size [18,19]. Extracellular diffusivity (D ex) was assumed to be hindered diffusion time so that Sex = exp (-b x Dex) [20]. MRI-derived cellularity (number of cells per unit area) was calculated as 2 x (3vin 2π ) 2 3 ⁄d 2 [19]. A previous study defined an unconventional “cellularity” as vin⁄(d x 100), with some clinical potential [12]. We defined the latter as the cellularity index to avoid any confusion with conventional cellularity. Additionally, ADC maps were obtained at each diffusion time according to S⁄S 0 = exp(-bD), where D is the diffusivity, using b = 250 s/mm2 for the 33-Hz OGSE data, b = 800 s/mm2 for the 17-Hz OGSE data, and b = 1,200 s/mm2 for the PGSE data to obtain diffusivity at 33 Hz (D33Hz), 17 Hz (D17Hz), and 0 Hz (D0Hz), respectively. Diffusion images were coregistered to the corresponding T2-weighted S (b = 0) images to correct for subject motion. Denoising employs a local principal component analysis filter [21]. Data fitting employed MATLAB R2024b (The MathWorks) to generate DWI parametric maps on a voxel- wise basis. Microstructural parameters (d, v in, and Dex) were fitted using the MATI package [22]. The fitting was repeated 100 times per sample, and the analysis with the smallest fitting residual was chosen as the final result. In patients with endometrial lesions, radiologist 1 (with Women with clinical suspicion of EC were enrolled to undergo MRI, including time-dependent diffusion MRI, between January 2024 and March 2025 (n = 177) Excluded • N o pathology confirmation (n = 22) • A typical endometrial hyperplasia (n = 13) • History o f prior treatment for endometrial disease before MRI (n = 3) • In sufficient MRI quality (n = 9) Benign lesions (n = 68) • U terine curettage (n = 48) • Sur gery (n = 20) EC (n = 62) • Sur gery (n = 62) 130 patients with complete time-dependent diffusion MRI sequence were included in the final analysis Fig. 1. Flowchart shows participant enrollment. EC = endometrial cancer 976 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org 10 years of experience in gynecological imaging) manually delineated the regions of interest (ROIs) on each slice of the lesions using PGSE DWI images with a b-value of 1,200 s/mm², carefully excluding the surrounding tissues. To enhance the representativeness and reduce potential measurement bias, the final value for each patient was calculated as the average across all lesion-containing slices. To further improve robustness, all delineations were independently resegmented by radiologist 2 (with 8 years of experience). The values used in the main analysis were the averages of the measurements taken by the two radiologists. Interobserver agreement between the two radiologists was assessed using intraclass correlation coefficients (ICCs) and Bland–Altman plots. Histopathological Analysis Biopsy specimens and histopathological slides were reviewed under the supervision of two experienced gynecological oncology pathologists with 15 and 10 years of experience. Benign endometrial lesions included endometrial polyps and hyperplasia. The pathological states of EC were recorded. According to the 2023 FIGO guidelines, low- grade (G1–G2) endometrioid endometrial carcinoma (EEC) is considered nonaggressive, whereas high-grade tumors, including EEC G3, serous carcinoma, clear cell carcinoma, and other rare subtypes, are classified as aggressive [3]. In our cohort, only EEC were present, with no non- endometrioid histological subtypes. Therefore, in accordance with the updated FIGO classification system, we defined EEC G3 as aggressive EC and EEC G1–G2 as nonaggressive EC. A B Fig. 2. Schematic shows the pulse sequences and differentiation of benign endometrial lesions and endometrial cancer using time- dependent diffusion magnetic resonance imaging-based microstructural mapping. A: The diagram presented illustrates the pulse sequences utilized for imaging microstructural parameters using a limited spectrally edited diffusion method. In addition to conventional PGSE ACQ, OGSE ACQ at two frequencies (n = 1 and 2) were employed. Diffusion signals, which are dependent on diffusion time, can be captured using both pulsed and OGSE diffusion encoding schemes across various diffusion times. B: The diffusivity of water molecules in a cellular environment is influenced by diffusion time, with this effect becoming more noticeable as cellular density increases. By employing pulsed and OGSE diffusion encoding schemes at varying diffusion times, diffusion signals can be captured. These signals enable the reconstruction of microstructural properties using biophysical modeling approaches. PGSE = pulsed gradient spin-echo, ACQ = acquisitions, OGSE = oscillating gradient spin-echo, td = diffusion time, NK = natural killer 977 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org Histopathological Assessment and Correlation With MRI Light microscopy histology images were analyzed using a deep learning-based cell pose method for cellular segmentation [23]. After all individual nuclei were identified, the histology-derived cellularity (the total number of nuclei divided by the total area of the section) was calculated. Fnuclei was calculated as the area fraction occupied by the nuclei. For each segmented nucleus, an effective diameter was calculated using d = (A/π) * 2, where A is the nuclear area. The mean nuclear diameter was calculated for each histological section to better correlate with the corresponding TDD-MRI-derived mean cell diameter. Statistical Analysis Continuous variables with normal distribution were compared using the two-sample t-test, and categorical variables using Pearson χ2 test and Fisher’s exact test. The differences in microstructural parameters between benign endometrial lesions and EC, as well as between ECs with different aggressiveness levels, were assessed using a two-sample t-test. Additionally, the Pearson correlation coefficient was used to assess the relationship between the microstructural parameters and histological indices. The interobserver agreement for the microstructural parameters was assessed using ICCs, with ICCs >0.75 indicating good agreement. Bland–Altman analysis was used to demonstrate the reproducibility of measurements of the microstructural parameters. Receiver operating characteristic (ROC) curve analysis was used to assess the diagnostic performance. To address the challenges posed by the small sample size and rarity of aggressive EC, we employed a Bayesian logistic regression model. This approach allows the incorporation of prior information and yields more stable estimates in the presence of limited data and rare events [24]. The model performance was evaluated using standard classification metrics, and the robustness of these performance measures was assessed by bootstrapping with 200 resampled datasets. We used the Youden index to determine the optimal threshold for classifying cases. The corresponding area under the ROC curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value were also calculated. Statistical analyses were performed using SPSS 26.0 (IBM Corp., Armonk, NY, USA), GraphPad Prism 8.0 (San Diego, CA, USA), MedCalc 22.0 (Oostende, Belgium) and R 4.1.0 (The R Project for Statistical Computing, a global open-source project). P < 0.05 indicated statistical significance.

Baseline Characteristics The final analysis included 130 women administered TDD-MRI followed by uterine curettage or surgery for whom complete clinical information was available. Their mean age was 56 ± 14 years, with an interquartile range of 45–66 years. Among all participants, 68 (52.3%) were pathologically diagnosed with benign endometrial lesions and 62 (47.7%) were diagnosed with EC and underwent staging surgery. Among these 62 patients, EC in 50 (80.6%) was identified as nonaggressive, whereas in 12 (19.4%), it was aggressive. The baseline characteristics of the participants are summarized in Table 1. Interobserver Agreement The ICCs for diameter, D ex, cellularity, cellularity index, vin, D0Hz, D17Hz and D33Hz were 0.93 (95% confidence interval [CI]: 0.90, 0.95), 0.96 (95% CI: 0.94, 0.97), 0.96 (95% CI: 0.94, 0.97), 0.96 (95% CI: 0.94, 0.97), 0.97 (95% CI: 0.96, 0.98), 0.94 (95% CI: 0.91, 0.96), 0.94 (95% CI: 0.92, 0.96), and 0.94 (95% CI: 0.91, 0.95) respectively. The Bland–Altman analysis showed good reproducibility in the measurement of microstructural parameters taken by the two radiologists (Fig. 3). Table 1. Patient baseline characteristics Clinical parameters Benign endometrial lesions (n = 68) EC (n = 62) Non-aggressive EC (n = 50) Aggressive EC (n = 12) P* P† Age, yr 52.96 ± 14.60 59.15 ± 11.97 58.46 ± 11.77 62.00 ± 12.90 0.010 0.362 BMI, kg/m2 25.16 ± 4.02 27.22 ± 5.25 27.59 ± 5.32 25.66 ± 4.82 0.013 0.256 Menopausal status, yes 32 (47.1) 43 (69.4) 35 (70.0) 8 (66.7) 0.010 0.822 Abnormal uterine bleeding, yes 29 (42.6) 48 (77.4) 38 (76.0) 10 (83.3) 0 0.585 Live birth, yes 53 (77.9) 54 (87.1) 44 (88.0) 10 (83.3) 0.172 0.665 Diabetes, yes 12 (17.6) 17 (27.4) 13 (26.0) 4 (33.3) 0.181 0.609 Data are presented as mean ± standard deviation or patient number with percentage in parentheses. *P-value for benign endometrial lesions and EC, †P-value for non-aggressive and aggressive EC. EC = endometrial cancer, BMI = body mass index 978 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org Comparing TDD-MRI-Derived Microstructural Parameters Between Different Endometrial Pathologies Figure 4 shows the T2-weighted images, TDD-MRI microstructural parameter maps, and diffusivity maps at different oscillating frequencies of benign endometrial lesions, nonaggressive EC, and aggressive EC. In terms of the comparison between benign endometrial lesions and EC, the microstructural parameters (diameter, D ex, cellularity, Fig. 3. Bland–Altman plots compare the reproducibility of time-dependent diffusion magnetic resonance imaging measurements from two independent readers. A: Plot shows that, for diameter, the mean difference is -0.08 μm (95% CI: -0.91, 0.75). B: Plot shows that, for Dex, the mean difference is 0.02 μm2/ms (95% CI: -0.13, 0.18). C: Plot shows that, for cellularity, the mean difference is 0.04 (95% CI: -0.36, 0.44). D: Plot shows that, for the cellularity index, the mean difference is -0.00 (95% CI: -0.35, 0.34). E: Plot shows that, for vin, the mean difference is 0 μm2/ms (95% CI: -0.04, 0.04). F: Plot shows that, for D0Hz, the mean difference is 0.01 μm2/ms (95% CI: -0.23, 0.25). G: Plot shows that, for D17Hz, the mean difference is 0.03 μm2/ms (95% CI: -0.23, 0.29). H: Plot shows that, for D33Hz, the mean difference is -0.02 μm2/ms (95% CI: -0.27, 0.23). The dotted line represents the mean difference, and the upper and lower solid lines represent the upper and lower limits of agreement, respectively. CI = confidence interval, Dex = extracellular diffusivity, vin = intracellular volume fraction, SD = standard deviation Fig. 4. Microstructural parameter maps of benign lesions and nonaggressive and aggressive endometrial cancer, including diameter, Dex, vin, cellularity, cellularity index and the diffusivity maps from pulsed gradient spin-echo (D0Hz) and oscillating gradient spin-echo (D17Hz and D33Hz) data. Corresponding T2W images at the similar axial locations are shown in the first column. Dex = extracellular diffusivity, vin, = intracellular volume fraction, T2W = T2-weighted, IMPULSED = Imaging Microstructural Parameters Using Limited Spectrally Edited Diffusion, ADC = apparent diffusion coefficient A E B F C G D H 979 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org cellularity index, and vin) showed significant differences (P < 0.001) (Fig. 5A-E). When comparing nonaggressive and aggressive EC, the diameter, Dex, cellularity, cellularity index, and vin showed significant differences (P < 0.05) (Fig. 5I-M). In benign endometrial lesions, Dex, cellularity, and cellularity index were significantly higher than those in EC, whereas diameter and vin were significantly lower. In nonaggressive EC, Dex, cellularity, and cellularity index were significantly higher than those in aggressive EC, whereas the diameter and v in were significantly lower. D0Hz, D17Hz, and D33Hz showed significant differences between benign endometrial lesions and EC (P < 0.05) (Fig. 5F-H), as well as between nonaggressive and aggressive EC (P < 0.05) (Fig. 5N-P). In benign endometrial lesions, D0Hz, D17Hz, and D33Hz were significantly higher than those in EC. In nonaggressive EC, D 0Hz, D17Hz and D33Hz were significantly higher than those in aggressive EC. Diagnostic Performance of TDD-MRI-Derived Microstructural Parameters In distinguishing between benign endometrial lesions and EC, cellularity achieved the highest performance among all microstructural features, with an AUC of 0.86 (95% CI: 0.79, 0.93), sensitivity of 74.2% (46 of 62 participants) and specificity of 91.2% (62 of 68 participants), followed by v in, with an AUC of 0.83 (95% CI: 0.75, 0.90), sensitivity of Fig. 5. Box and whisker plots show comparisons of microstructural parameters, including (A) diameter, (B) Dex, (C) cellularity, (D) cellularity index, (E) vin, (F) D0Hz, (G) at D17Hz, and (H) at D33Hz among benign lesions and EC, (I) diameter, (J) Dex, (K) cellularity, (L) cellularity index, (M) vin, (N) at D0Hz, (O) at D17Hz, and (P) at D33Hz among nonaggressive EC and aggressive EC. *P < 0.05, †P < 0.01, ‡P < 0.001. Dots represent individual data points, boxes indicate the standard deviation, and midlines are the median. Dex = extracellular diffusivity, vin = intracellular volume fraction, EC = endometrial cancer A E I M B F J N C G K O D H L P ‡ ‡ ‡‡ ‡ ‡ ‡ ‡ ‡ † † † † 980 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org 93.5% (58 of 62 participants) and specificity of 66.2% (45 of 68 participants) (Table 2). In distinguishing between nonaggressive and aggressive EC, the classification analysis revealed that, among all microstructural features, the cellularity index achieved the highest performance, with an AUC of 0.88 (95% CI: 0.78, 0.95), sensitivity of 100.0% (12 of 12 participants) and specificity of 70.0% (35 of 50 participants), followed by diameter, with an AUC of 0.84 (95% CI: 0.70, 0.93), sensitivity of 91.7% (11 of 12 participants) and specificity of 66.0% (33 of 50 participants) (Table 3). Correlation Between TDD-MRI-Derived Microstructural Parameters and D 0Hz Figure 6 shows the correlations between the fitted microstructural parameters and conventionally used D0Hz at the participant level. D0Hz was positively correlated with Dex (P < 0.05) and negatively correlated with diameter (P < 0.05), cellularity index (P < 0.01) and vin (P < 0.001) in benign endometrial lesions. D0Hz was positively correlated with Dex (P < 0.001) and negatively correlated with vin (P < 0.001) in patients with EC. D0Hz was positively correlated with Dex (P < 0.05) and negatively correlated with vin (P < 0.001) in nonaggressive EC. D0Hz was positively correlated with Dex (P < 0.05) and negatively correlated with cellularity ( P < 0.01) in aggressive EC. Correlation and Simulation Validation With Histopathological Findings Cell nuclei were automatically segmented from hematoxylin-eosin-stained whole-slide images (Fig. 7A-C). Histopathological correlation analysis was conducted in a representative subset of 16 patients who underwent surgery, including 6 with benign endometrial lesions and 10 with EC. Surgical specimens were specifically chosen to ensure consistent tissue sampling and high-quality hematoxylin- eosin-stained slides, as they offer more standardized and comprehensive histological material, compared with curettage samples. Among the patients, the EC in six was classified as aggressive subtypes. (v in, r = 0.768; diameter, r = 0.768; cellularity, r = 0.832; all P < 0.001) (Fig. 7D-F).

This study prospectively assessed patients with endometrial diseases to demonstrate the clinical application of microstructural parameters obtained using TDD-MRI for differentiating benign and malignant endometrial diseases and classifying aggressiveness of EC. These microstructural parameters demonstrated excellent predictive performance in differentiating endometrial diseases, with cellularity achieving the best performance, with an AUC of 0.86. For distinguishing EC aggressiveness levels, the cellularity Table 2. Diagnostic performance of time-dependent diffusion MRI–derived microstructural parameters for distinguishing benign and cancerous endometrial lesions Characteristic AUC Cutoff Sensitivity, % Specificity, % PPV, % NPV, % Cellularity, cells x 10-3/μm2 0.86 [0.79, 0.93] 0.55 74.2 (46/62) [63.3, 85.1] 91.2 (62/68) [84.4, 97.9] 88.5 (46/52) [79.8, 97.1] 79.5 (62/78) [70.5, 88.4] Vin 0.83 [0.75, 0.90] 0.38 93.5 (58/62) [89.8, 100] 66.2 (45/68) [54.9, 77.4] 71.6 (58/81) [62.2, 81.7] 91.8 (45/49) [86.9, 100] Diameter, μm 0.81 [0.74, 0.88] 0.31 96.8 (60/62) [92.4, 100] 58.8 (40/68) [47.1, 70.5] 68.2 (60/88) [58.5, 77.9] 95.2 (40/42) [88.8, 100] Dex, μm2/msec 0.74 [0.65, 0.82] 0.42 85.5 (53/62) [76.7, 94.3] 60.3 (41/68) [48.7, 71.9] 66.2 (53/80) [55.9, 76.6] 82.0 (41/50) [71.4, 92.6] Diffusivity at 0 Hz, μm2/msec 0.71 [0.62, 0.80] 0.48 71.0 (44/62) [59.7, 82.3] 69.1 (47/68) [58.1, 80.1] 67.7 (44/65) [56.3, 79.1] 72.3 (47/65) [61.4, 83.2] Cellularity index, μm-1 0.70 [0.60, 0.79] 0.43 95.2 (59/62) [89.8, 100] 57.4 (39/68) [45.6, 69.1] 67.0 (59/88) [57.2, 76.9] 92.9 (39/42) [85.1, 100] Diffusivity at 17 Hz, μm2/msec 0.67 [0.58, 0.77] 0.48 67.7 (42/62) [56.1, 79.4] 69.1 (47/68) [58.1, 80.1] 66.7 (42/63) [55.0, 78.3] 70.1 (47/67) [59.2, 81.1] Diffusivity at 33 Hz, μm2/msec 0.62 [0.53, 0.72] 0.47 61.3 (38/62) [49.2, 73.4] 61.8 (42/68) [50.2, 73.3] 59.4 (38/64) [47.3, 71.4] 63.6 (42/66) [52.0, 75.2] The values in parentheses are numerators and denominators, and the values in brackets are 95% confidence intervals. AUC = area under the receiver operating characteristic curve, PPV = positive predictive value, NPV = negative predictive value, vin = intracellular volume fraction, Dex = extracellular diffusivity 981 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org index achieved the best performance, with an AUC of 0.88. The microstructural parameters derived from TDD- MRI showed strong correlation with pathology-determined microstructural properties (r = 0.77–0.83; P < 0.001). Accurate, noninvasive preoperative diagnosis of endometrial lesions and assessment of aggressiveness of EC are crucial for determining appropriate clinical management. Although benign and malignant endometrial conditions often have similar symptoms, their treatment strategies differ significantly. Benign lesions may not require surgical intervention, whereas malignant lesions typically necessitate hysterectomy with bilateral salpingo- oophorectomy and lymph node assessment. Although endometrial sampling via hysteroscopy or curettage remains the standard method for histopathological confirmation, its diagnostic accuracy can be limited by the lesion location, sampling limitations, and/or operator experience. Moreover, in patients with cervical stenosis, poor surgical tolerance, or Table 3. Diagnostic performance of time-dependent diffusion MRI–derived microstructural parameters in distinguishing non-aggressive and aggressive endometrial cancers Characteristic AUC (95% CI) Cutoff (95% CI) Sensitivity, % Specificity, % PPV, % NPV, % Cellularity index, μm-1 0.88 [0.78, 0.95] 0.17 100 (12/12) [76.9, 100] 70.0 (35/50) [58.0, 96.2] 44.4 (12/27) [28.6, 85.8] 100 (35/35) [94.0, 100] Diameter, μm 0.84 [0.70, 0.93] 0.11 91.7 (11/12) [60.0, 100] 66.0 (33/50) [44.6, 96.2] 39.3 (11/28) [23.1, 84.6] 97.1 (33/34) [89.1, 100] Diffusivity at 33 Hz, μm2/msec 0.79 [0.65, 0.90] 0.11 100 (12/12) [54.5, 100] 50.0 (25/50) [40.9, 98.1] 32.4 (12/37) [21.4, 87.6] 100 (25/25) [88.0, 100] Vin 0.79 [0.62, 0.93] 0.16 83.3 (10/12) [57.1, 100] 82.0 (41/50) [74.4, 94.4] 52.6 (10/19) [33.3, 75.0] 95.3 (41/43) [88.6, 100] Diffusivity at 0 Hz, μm2/msec 0.76 [0.56, 0.92] 0.18 83.3 (10/12) [38.4, 100] 62.0 (31/50) [45.3, 100] 34.5 (10/29) [21.6, 100] 93.9 (31/33) [83.0, 100] Dex, μm2/msec 0.74 [0.50, 0.92] 0.15 83.3 (10/12) [44.4, 100] 56.0 (28/50) [28.5, 100] 31.2 (10/32) [19.0, 100] 93.3 (28/30) [83.0, 100] Diffusivity at 17 Hz, μm2/msec 0.66 [0.40, 0.85] 0.16 75.0 (9/12) [30.7, 100] 54.0 (27/50) [24.4, 100] 28.1 (9/32) [17.6, 100] 90.0 (27/30) [81.0, 100] Cellularity, cells x 10-3/μm2 0.61 [0.46, 0.86] 0.36 [0.09, 0.53] 50.0 (6/12) [36.3, 83.4] 100 (50/50) [65.8, 100] 100 (6/6) [20.0, 100] 89.3 (50/56) [81.1, 96.4] All diagnostic performance metrics (AUC, sensitivity, specificity, PPV, and NPV) are estimated using posterior mean from Bayesian logistic regression model with weakly informative prior N (0, 2.52). 95% CIs were calculated using Bootstrapping with 200 resampled datasets. The values in parentheses are numerators and denominators, and the values in brackets are 95% CIs. AUC = area under the receiver operating characteristic curve, PPV = positive predictive value, NPV = negative predictive value, CI = confidence interval, v in = intracellular volume fraction, Dex = extracellular diffusivity Fig. 6. Correlations between the fitted microstructural parameters and D0Hz at the participant level (A-E) among benign lesions and EC and (F-J) among nonaggressive and aggressive EC. *P < 0.05, †P < 0.01, ‡P < 0.001. EC = endometrial cancer A F B G C H D I E J ‡ † † ‡ ‡ † 982 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org other anatomical constraints, these invasive procedures may be contraindicated or pose significant risk [25]. By enhancing diagnostic accuracy in challenging or indeterminate cases, MRI reduces the need for unnecessary biopsies or repeat sampling. A more accurate, noninvasive assessment of lesion type and tumor aggressiveness could inform clinical decisions regarding whether to proceed with surgery, consider a more conservative approach, or explore fertility-preserving strategies for younger patients. Ultimately, the integration of MRI into the diagnostic workflow may improve preoperative risk stratification, support personalized treatment planning, avoid overtreatment, and reduce patient burden. Multi-parametric MRI, especially DWI and ADC, is indispensable in oncology. However, ADC only measures water diffusivity, which is determined by various microstructural features [26-28]. ADC accuracy is affected by intracellular and extracellular diffusion coefficients, intracellular volume fraction, capillary perfusion, cell membrane permeability, and cell size [29]. As changes in tumor cell size reflect both disease status and therapeutic response, the ability to noninvasively assess tissue microstructure is essential for monitoring tumor behavior and treatment outcomes. Previous studies have explored the ability of functional molecular MRI and advanced diffusion MRI techniques to differentiate the histological information of endometrial diseases and EC [30-32]. However, some parameters in the aforementioned studies exhibited relatively limited predictive performance. TDD- MRI, particularly with OGSE acquisitions to broaden the diffusion time range, provides a unique means to gain sensitivity to different length scales [10], enabling the characterization of different microstructures in tumors and clinical feasibility to differentiate tumor malignancy and grades [33-35]. There is increasing interest in TDD- MRI, which uses multi-b values, multi-diffusion times, and multi-compartmental biophysical models to quantitatively characterize microstructural information at the cellular level. Therefore, IMPULSED has been rapidly implemented in several clinical cancers, and the derived microstructural features have promising potential for clinical applications such as differentiating between clinically significant and insignificant prostate cancers [12,13,15] and predicting the treatment response in breast cancer [14]. Compared with previous studies, our study implemented TDD-MRI to differentiate endometrial diseases and EC and further classify aggressiveness of EC. Additionally, we conducted a consistency test on the microstructural parameters from the ROI measurements taken by two radiologists, and the ICC was >0.90, which demonstrated good stability of the measurements. A D B E C F Fig. 7. Correlations between time-dependent diffusion MRI-derived microstructural parameters and pathology-based microstructural features (n = 16). A: Hematoxylin-eosin-stained image (x40) shows pathological specimen from one participant with benign lesion. B: Hematoxylin-eosin-stained image (x40) illustrates nuclei segmented by a pretrained conditional generative adversarial network. C: Automated quantification of the pathological microstructural features. D-F: The graph depicts the correlations between time- dependent diffusion MRI-derived (D) diameter, (E) cellularity, (F) v in, and the pathology-based microstructural features. fnuclei = nuclei fraction, vin = intracellular volume fraction 983 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org As shown above, compared with benign endometrial diseases, EC exhibits higher cell diameters and vin and lower cellularity, cellularity index, and Dex. Additionally, the ADC values across the three oscillation frequencies were lower in EC, reflecting the restricted diffusion typically observed in dense tissues, including tumors, lymph nodes, or fibrotic regions [36]. Malignant tumor cells usually demonstrate larger nuclei, greater pleomorphism, and higher proliferative activity, compared with normal cells [4], which may lead to reduced cellularity per unit area and a corresponding decrease in the cellularity index. These findings were supported by our pathological analysis. Moreover, aggressive EC exhibits higher cell diameters and v in values, compared with nonaggressive EC, while displaying lower cellularity, cellularity index, and D ex. Furthermore, the ADC values of aggressive EC were lower than those of nonaggressive EC. These features are consistent with a more disorganized and heterogeneous tumor microenvironment, where rapid proliferation may outpace neovascularization, resulting in relatively fewer viable tumor cells per unit area and more necrotic zones [37]. The difference in the cellularity index between aggressive and nonaggressive EC is more pronounced. From the perspective of calculation, the cellularity index tends to show less fluctuation and is thus less susceptible to noise; yet its overall trend is consistent with that of cellularity, both being lower in aggressive EC. Importantly, the pathological features of EC (e.g., cellular atypia and necrosis) may influence the TDD-MRI parameters. Increased atypia enhances intracellular complexity and reduces the extracellular space, boosting restricted diffusion and decreasing the ADC or D ex values. Conversely, necrosis may increase diffusivity due to the loss of structural barriers and increase free water content, potentially contributing to localized heterogeneity in DWI-derived parameters. v in is significantly underestimated owing to transcytolemmal water exchange [38] but strongly correlates with pathology- derived values [19], indicating its potential utility in reflecting microstructural changes. In patients with EC, D ex and vin values obtained via TDD-MRI were correlated with D0Hz. Further, Dex and vin were correlated with D0Hz in nonaggressive EC. Dex and cellularity were correlated with D0Hz in nonaggressive EC, corroborating previous studies [14]. Our study has several limitations. The sample size was relatively limited, particularly in certain subgroups, such as aggressive EC, which accounts for only one-sixth of all EC cases, according to epidemiological data. Although our cohort reflects the disease distribution, the small number of aggressive cases may limit the statistical power for subgroup comparisons. To address this limitation, we plan to expand patient recruitment in future studies, focusing on the underrepresented subtypes. A larger cohort with improved control of confounding factors will enable a more comprehensive molecular-level evaluation of EC and more robust statistical analyses. Second, the microstructural parameters were based on the ROI averages. Future work will explore advanced postprocessing to visualize MRI microstructural parameters across lesions and the surrounding environment. Furthermore, combining TDD- MRI with other novel MRI technologies to develop models incorporating multiple imaging modalities and parameters can enhance our ability to noninvasively predict molecular- level information related to endometrial diseases and EC. In summary, this study demonstrated the feasibility and high diagnostic performance of TDD-MRI-derived microstructural parameters in non-invasively differentiating benign and malignant endometrial diseases, as well as in identifying the aggressiveness of EC. Further validation is warranted, with a larger sample size, multicenter collaboration, and incorporation of various pathological and molecular information. Availability of Data and Material The datasets generated or analyzed during the study are available from the corresponding author on reasonable request. Conflicts of Interest The authors have no potential conflicts of interest to disclose. Author Contributions Conceptualization: Qi Yang, Hua Li, Junzhong Xu. Data curation: Wenyi Yue, Ruxue Han, Junzhong Xu. Formal analysis: Wenyi Yue, Ruxue Han, Ning Xu. Funding acquisition: Qi Yang, Hua Li, Dan Zhao. Investigation: Wenyi Yue, Ruxue Han. Methodology: Dandan Zheng, Jing Peng. Project administration: Qi Yang, Hua Li, Dan Zhao. Resources: Chaoyang Jin, Xiaoyu Jiang. Software: Junzhong Xu, Chaoyang Jin, Xiaoyu Jiang. Supervision: Qiming Liu, Ning Xu. Validation: Jun Lu, Chaoyang Jin, Xiaoyu Jiang. Visualization: Wenyi Yue, Ruxue Han, Junzhong Xu, Dandan Zheng. Writing—original draft: Wenyi Yue, Ruxue Han, Junzhong Xu. Writing—review & editing: Qi Yang, Hua Li. 984 Yue et al. https://doi.org/10.3348/kjr.2025.0633 kjronline.org ORCID IDs Wenyi Yue https://orcid.org/0009-0005-1104-5153 Ruxue Han https://orcid.org/0009-0005-0577-7286 Junzhong Xu https://orcid.org/0000-0001-7895-4232 Chaoyang Jin https://orcid.org/0000-0002-9749-8962 Xiaoyu Jiang https://orcid.org/0000-0002-0369-6301 Dandan Zheng https://orcid.org/0000-0001-8526-0386 Jing Peng https://orcid.org/0000-0002-8835-4545 Jun Lu https://orcid.org/0000-0003-0363-7288 Qiming Liu https://orcid.org/0000-0002-7928-4557 Ning Xu https://orcid.org/0000-0002-7770-8731 Dan Zhao https://orcid.org/0000-0002-3231-267X Hua Li https://orcid.org/0000-0002-1172-6750 Qi Yang https://orcid.org/0000-0002-5773-0456 Funding Statement This work was supported by the Beijing Hospitals Authority’s Ascent Plan (DFL20220303), Beijing Key Specialists in Major Epidemic Prevention and Control, Beijing Science and Technology Cross-disciplinary New Star Project, National Natural Science Foundation of China (62476180), National Natural Science Foundation of China (62176267), CAMS Innovation Fund for Medical Sciences (2024-I2M- C&T-A-003), Qinghai Provincial Science and Technology Plan Special Aid Project for Qinghai (2025-QY-220).

1. Lefebvre TL, Ueno Y, Dohan A, Chatterjee A, Vallières M, Winter-Reinhold E, et al. Development and validation of multiparametric MRI-based radiomics models for preoperative risk stratification of endometrial cancer. Radiology 2022;305:375-386 2. Clarke MA, Long BJ, Del Mar Morillo A, Arbyn M, Bakkum- Gamez JN, Wentzensen N. Association of endometrial cancer risk with postmenopausal bleeding in women: a systematic review and meta-analysis. JAMA Intern Med 2018;178:1210- 1222 3. Berek JS, Matias-Guiu X, Creutzberg C, Fotopoulou C, Gaffney D, Kehoe S, et al. FIGO staging of endometrial cancer: 2023. J Gynecol Oncol 2023;34:e85 4. Yu Z, Zhang J, Zhang Q, Wei S, Shi R, Zhao R, et al. Single- cell sequencing reveals the heterogeneity and intratumoral crosstalk in human endometrial cancer. Cell Prolif 2022;55:e13249 5. Ren F, Wang L, Wang Y, Wang J, Wang Y, Song X, et al. Single- cell transcriptome profiles the heterogeneity of tumor cells and microenvironments for different pathological endometrial cancer and identifies specific sensitive drugs. Cell Death Dis 2024;15:571 6. Ren X, Liang J, Zhang Y, Jiang N, Xu Y, Qiu M, et al. Single- cell transcriptomic analysis highlights origin and pathological process of human endometrioid endometrial carcinoma. Nat Commun 2022;13:6300 7. Horn LC, Meinel A, Handzel R, Einenkel J. Histopathology of endometrial hyperplasia and endometrial carcinoma: an update. Ann Diagn Pathol 2007;11:297-311 8. Subramaniam KS, Tham ST, Mohamed Z, Woo YL, Mat Adenan NA, Chung I. Cancer-associated fibroblasts promote proliferation of endometrial cancer cells. PLoS One 2013;8:e68923 9. Fujii S, Gonda T, Yunaga H. Clinical utility of diffusion-weighted imaging in gynecological imaging: revisited. Invest Radiol 2024;59:78-91 10. Gore JC, Xu J, Colvin DC, Yankeelov TE, Parsons EC, Does MD. Characterization of tissue structure at varying length scales using temporal diffusion spectroscopy. NMR Biomed 2010;23:745-756 11. Xu J, Jiang X, Li H, Arlinghaus LR, McKinley ET, Devan SP, et al. Magnetic resonance imaging of mean cell size in human breast tumors. Magn Reson Med 2020;83:2002-2014 12. Wu D, Jiang K, Li H, Zhang Z, Ba R, Zhang Y, et al. Time- dependent diffusion MRI for quantitative microstructural mapping of prostate cancer. Radiology 2022;303:578-587 13. Ba R, Wang X, Zhang Z, Li Q, Sun Y, Zhang J, et al. Diffusion- time dependent diffusion MRI: effect of diffusion-time on microstructural mapping and prediction of prognostic features in breast cancer. Eur Radiol 2023;33:6226-6237 14. Wang X, Ba R, Huang Y, Cao Y, Chen H, Xu H, et al. Time- dependent diffusion MRI helps predict molecular subtypes and treatment response to neoadjuvant chemotherapy in breast cancer. Radiology 2024;313:e240288 15. Zhang H, Liu K, Ba R, Zhang Z, Zhang Y, Chen Y, et al. Histological and molecular classifications of pediatric glioma with time-dependent diffusion MRI-based microstructural mapping. Neuro Oncol 2023;25:1146-1156 16. Cao Y, Lu Y, Shao W, Zhai W, Song J, Zhang A, et al. Time- dependent diffusion MRI-based microstructural mapping for differentiating high-grade serous ovarian cancer from serous 985 Time-Dependent Diffusion MRI of Endometrial Cancer https://doi.org/10.3348/kjr.2025.0633kjronline.org borderline ovarian tumor. Eur J Radiol 2024;178:111622 17. Solomon E, Lemberskiy G, Baete S, Hu K, Malyarenko D, Swanson S, et al. Time-dependent diffusivity and kurtosis in phantoms and patients with head and neck cancer. Magn Reson Med 2023;89:522-535 18. Jiang X, Li H, Xie J, Zhao P, Gore JC, Xu J. Quantification of cell size using temporal diffusion spectroscopy. Magn Reson Med 2016;75:1076-1085 19. Jiang X, Li H, Xie J, McKinley ET, Zhao P, Gore JC, et al. In vivo imaging of cancer cell size and cellularity using temporal diffusion spectroscopy. Magn Reson Med 2017;78:156-164 20. Reynaud O, Winters KV, Hoang DM, Wadghiri YZ, Novikov DS, Kim SG. Pulsed and oscillating gradient MRI for assessment of cell size and extracellular space (POMACE) in mouse gliomas. NMR Biomed 2016;29:1350-1363 21. Manjón JV, Coupé P, Concha L, Buades A, Collins DL, Robles M. Diffusion weighted image denoising using overcomplete local PCA. PLoS One 2013;8:e73021 22. Xu J, Devan SP, Shi D, Pamulaparthi A, Yan N, Zu Z, et al. MATI: a GPU-accelerated toolbox for microstructural diffusion MRI simulation and data fitting with a graphical user interface. Magn Reson Imaging 2025;122:110428 23. Stringer C, Wang T, Michaelos M, Pachitariu M. Cellpose: a generalist algorithm for cellular segmentation. Nat Methods 2021;18:100-106 24. Gelman A, Jakulin A, Pittau MG, Su YS. A weakly informative default prior distribution for logistic and other regression models. Ann Appl Stat 2008;2:1360-1383 25. Helpman L, Kupets R, Covens A, Saad RS, Khalifa MA, Ismiil N, et al. Assessment of endometrial sampling as a predictor of final surgical pathology in endometrial cancer. Br J Cancer 2014;110:609-615 26. Chatterjee A, Watson G, Myint E, Sved P, McEntee M, Bourne R. Changes in epithelium, stroma, and lumen space correlate more strongly with Gleason pattern and are stronger predictors of prostate ADC changes than cellularity metrics. Radiology 2015;277:751-762 27. Meyer HJ, Wienke A, Surov A. ADC values of benign and high grade meningiomas and associations with tumor cellularity and proliferation - A systematic review and meta-analysis. J Neurol Sci 2020;415:116975 28. Wu X, Pertovaara H, Dastidar P, Vornanen M, Paavolainen L, Marjomäki V, et al. ADC measurements in diffuse large B-cell lymphoma and follicular lymphoma: a DWI and cellularity study. Eur J Radiol 2013;82:e158-e164 29. Novikov DS. The present and the future of microstructure MRI: from a paradigm shift to normal science. J Neurosci Methods 2021;351:108947 30. Tian S, Chen A, Li Y, Wang N, Ma C, Lin L, et al. The combined application of amide proton transfer imaging and diffusion kurtosis imaging for differentiating stage Ia endometrial carcinoma and endometrial polyps. Magn Reson Imaging 2023;99:67-72 31. Ma C, Tian S, Song Q, Chen L, Meng X, Wang N, et al. Amide proton transfer-weighted imaging combined with intravoxel incoherent motion for evaluating microsatellite instability in endometrial cancer. J Magn Reson Imaging 2023;57:493-505 32. Yamada I, Wakana K, Kobayashi D, Miyasaka N, Oshima N, Wakabayashi A, et al. Endometrial carcinoma: evaluation using diffusion-tensor imaging and its correlation with histopathologic findings. J Magn Reson Imaging 2019;50:250-260 33. Xu J, Li K, Smith RA, Waterton JC, Zhao P, Chen H, et al. Characterizing tumor response to chemotherapy at various length scales using temporal diffusion spectroscopy. PLoS One 2012;7:e41714 34. Iima M, Yamamoto A, Kataoka M, Yamada Y, Omori K, Feiweier T, et al. Time-dependent diffusion MRI to distinguish malignant from benign head and neck tumors. J Magn Reson Imaging 2019;50:88-95 35. Maekawa T, Hori M, Murata K, Feiweier T, Kamiya K, Andica C, et al. Differentiation of high-grade and low-grade intra-axial brain tumors by time-dependent diffusion MRI. Magn Reson Imaging 2020;72:34-41 36. Bollineni VR, Kramer G, Liu Y, Melidis C, deSouza NM. A literature review of the association between diffusion-weighted MRI derived apparent diffusion coefficient and tumour aggressiveness in pelvic cancer. Cancer Treat Rev 2015;41:496-502 37. Tamai K, Koyama T, Saga T, Umeoka S, Mikami Y, Fujii S, et al. Diffusion-weighted MR imaging of uterine endometrial cancer. J Magn Reson Imaging 2007;26:682-687 38. Li H, Jiang X, Xie J, Gore JC, Xu J. Impact of transcytolemmal water exchange on estimates of tissue microstructural properties derived from diffusion MRI. Magn Reson Med 2017;77:2239-2249