Time-dependent diffusion magnetic resonance imaging can distinguish between two molecular subtypes of endometrial cancer before any tissue is taken, according to a prospective single-center study published in European Radiology Experimental. Researchers found that microstructural parameters derived from this advanced form of diffusion-weighted MRI differed significantly between tumors with deficient mismatch repair and those with proficient mismatch repair, and that combining two of these parameters achieved an area under the receiver operating characteristic curve of 0.903, outperforming conventional apparent diffusion coefficient measurements.

Mismatch repair proteins, including MLH1, PMS2, MSH2, and MSH6, allow cells to correct errors that arise during DNA replication. When at least one of these proteins is lost, tumors are classified as mismatch repair deficient; intact expression of all four defines mismatch repair proficient status. Deficient tumors represent approximately 20 to 30 percent of all endometrial cancers and carry distinct therapeutic implications. They show increased expression of programmed death-ligand 1, and clinical trials such as KEYNOTE-158 and GARNET have demonstrated durable anti-tumor activity of immunotherapies including pembrolizumab and dostarlimab in this subtype. Deficient mismatch repair is also present in over 90 percent of endometrial cancer patients with Lynch syndrome, making its detection important for cancer screening as well as treatment planning.

Currently, determining mismatch repair status relies on immunohistochemistry performed on biopsy or surgical specimens, a multistep, labor-intensive process that depends on tissue availability and can be affected by tumor heterogeneity and sampling error. Because MRI is already a routine part of preoperative assessment for endometrial cancer, a method that reads molecular status directly from imaging could offer a time-sensitive advantage. Conventional diffusion-weighted imaging uses a pulsed gradient spin-echo sequence with diffusion times typically longer than 20 milliseconds, which limits its sensitivity to cellular-scale structure. The newer time-dependent approach employs oscillating gradient spin-echo sequences at 20 and 40 hertz, reducing diffusion times to 12.73 and 6.36 milliseconds respectively, allowing the signal to probe finer details of the tissue microenvironment.

The study enrolled 131 consecutive patients with suspicious endometrial pathology between September 2024 and June 2025 at a single center. After exclusions for poor image quality or lesions smaller than 20 square millimeters per slice, 82 patients remained in the final cohort: 17 with mismatch repair deficient tumors, aged 61.8 ± 10.1 years, and 65 with proficient tumors, aged 58.6 ± 11.8 years. All participants underwent preoperative pelvic MRI on 3.0-Tesla scanners using both pulsed gradient and oscillating gradient diffusion sequences, with the full diffusion-weighted protocol taking approximately eight minutes. Tumor status was confirmed by immunohistochemistry on surgically resected specimens obtained within one month of imaging, examined by a board-certified pathologist with over ten years of experience.

Researchers fitted the imaging data using three biophysical models: IMPULSED, which treats tumor cells as impermeable spheres and ignores water exchange; JOINT and EXCHANGE, which both incorporate transcytolemmal water exchange into the modeling and yield an additional parameter, the water exchange rate constant. The derived parameters included intracellular volume fraction, cell diameter, cellularity calculated as intracellular volume fraction divided by cell diameter, extracellular diffusivity, and the exchange rate constant. A trained postgraduate student manually segmented the entire tumor on diffusion-weighted images, with volumes of interest verified by a board-certified radiologist with more than 15 years of pelvic imaging experience; both were blinded to clinicopathological information beyond the diagnosis of endometrial cancer itself.

The results showed a consistent pattern across all three modeling methods. Mismatch repair deficient tumors exhibited significantly higher intracellular volume fraction and cellularity, and significantly lower extracellular diffusivity and apparent diffusion coefficient values, compared with proficient tumors, with p values of 0.003 or lower. No significant differences emerged in cell diameter across the three methods, nor in the water exchange rate constant derived from the exchange-sensitive models. Inter-reader intraclass correlation coefficients ranged from 0.852 to 0.982 and intra-reader values from 0.963 to 0.990, indicating good to excellent reliability for all measured parameters.

Corroborating the imaging findings against tissue, the researchers digitized hematoxylin and eosin-stained whole-slide images from a subset of 23 patients and counted cell nuclei using dedicated software. Histological cell density correlated positively with the imaging-derived cellularity from the IMPULSED method, with a Pearson correlation coefficient of 0.771 and a p value below 0.001 in this subset, which included six deficient and seventeen proficient tumors. This direct histopathological correlation supports the biological plausibility of the imaging measurements, although the authors note that whole-slide digitization was not part of routine pathology workflow at their institution and was performed retrospectively only for slides meeting quality requirements.

Diagnostic performance varied by parameter and model. Using IMPULSED, intracellular volume fraction alone achieved an area under the curve of 0.831; with JOINT, the same parameter reached 0.838 and cellularity 0.821; with EXCHANGE, intracellular volume fraction and extracellular diffusivity scored 0.824 and 0.815 respectively. Combining intracellular volume fraction with extracellular diffusivity improved discrimination over conventional ADC in all cases: the IMPULSED combination raised the area under the curve to 0.908, a significant improvement with a DeLong test p value of 0.025, while the JOINT combination reached 0.903 with a p value of 0.046. The EXCHANGE combination improved to 0.876 but without statistical significance. Bootstrap internal validation with 1,000 iterations showed limited optimism across all three combination models, and a sensitivity analysis confirmed that scanner type was not a significant predictor in any model.

The researchers selected JOINT as the preferred combination model based on superior parametric precision for extracellular diffusivity, reflected in a lower standard error of the log odds ratio compared with IMPULSED, and marginally better calibration. Interestingly, incorporating transcytolemmal water exchange did not improve discrimination of mismatch repair status, which the authors attribute to the limited sensitivity to exchange at the oscillating gradient frequencies achievable on current clinical scanners, where effective diffusion times remain considerably shorter than the intracellular water residence time. They also note that the estimated exchange rate constant may be confounded by perfusion and intracellular-intravascular water exchange, introducing systematic bias that attenuates its discriminative sensitivity.

The biological interpretation links the imaging signature to known features of the tumor microenvironment. Deficient mismatch repair tumors are characterized by dense tumor-infiltrating lymphocyte infiltration and stromal remodeling, producing a more heterogeneous and denser cellular environment than proficient tumors. Impaired DNA repair and genomic instability increase neoantigen load, rendering tumor cells more prone to attack by lymphocytes and other immune cells, and prior studies have reported higher expression levels of T cells, macrophages, and natural killer cells in deficient tumors. Higher cellularity would in turn compress the extracellular space, consistent with the lower extracellular diffusivity observed. The authors emphasize, however, that these specific cellular and matrix mechanisms were not directly assessed in the present study and remain hypothesis-generating, requiring dedicated histological or immunohistochemical confirmation. They also caution that the extracellular compartment in the current models includes the intravascular space, so extracellular diffusivity reflects combined effects of extracellular and vascular changes.

Several limitations warrant cautious interpretation. The 20 square millimeter lesion threshold, constrained by spatial resolution, may bias the cohort toward larger tumors. The relatively small sample size, class imbalance with only 17 deficient cases, and slightly low events per variable limit statistical power, and conventional clinicopathological features did not differ significantly between groups, though these comparisons were underpowered. Both scanners came from a single vendor with similar acquisition protocols, and the single-center design with preoperatively recruited patients may restrict generalizability across histological subtypes and demographics. Manual tumor delineation is time-consuming and reader-dependent, and modeling cells as ideal spheres represents a fundamental approximation, as in vivo tissues contain multiple cell populations with heterogeneous sizes and non-spherical morphologies. The statistical analysis used volume-of-interest or patient-wise mean values, not accounting for intra-tumor heterogeneity. External validation in independent, prospective, multicenter, multi-vendor cohorts remains necessary before clinical implementation.

Nevertheless, the study demonstrates that time-dependent diffusion MRI can noninvasively capture microstructural differences tied to mismatch repair biology in endometrial cancer. The joint oscillating-pulsed gradient protocol is available on a commercial sharing platform and runs in approximately eight minutes on clinically available 3.0-Tesla scanners, with preprocessing and model fitting executable on a standard personal computer using the open-source MATI software package. The authors conclude that the derived microstructural parameters, intracellular volume fraction and extracellular diffusivity, could potentially serve as a useful method to improve the diagnostic performance of diffusion-weighted MRI in determining mismatch repair status, but further external validation is required before these findings can inform routine practice.

Subject of Research: Time-dependent diffusion MRI microstructural parameters for noninvasive preoperative determination of mismatch repair status in endometrial cancer

Article Title: Time-dependent diffusion MRI in determining mismatch repair status of endometrial cancer

Article References: Zhang, J., Shi, D., Li, J. X. J., Hwang, C. N., Liu, F., Wang, Z., Ho, G., Feiweier, T., Wei, J., Guo, H., & Lee, E. Y. (2026). Time-dependent diffusion MRI in determining mismatch repair status of endometrial cancer. European Radiology Experimental, 10(1), Article 167. https://doi.org/10.1186/s41747-026-00830-y

Image Credits: AI Generated

DOI: 10.1186/s41747-026-00830-y

Keywords: endometrial cancer, time-dependent diffusion MRI, mismatch repair, microstructural parameters, immunohistochemistry, diffusion-weighted imaging, OGSE, tumor microenvironment, diagnostic performance, gynecological oncology, biophysical modeling, noninvasive imaging