Virtual MR elastography and continuous-time random walk for predicting Ki-67 status in cervical cancer: a preliminary...

This study aimed to evaluate the performance of virtual magnetic resonance elastography (vMRE) and continuous-time random walk (CTRW) models in predicting the Ki-67 status of cervical cancer (CC).

A total of 85 CC patients were enrolled. Moreover, parameters derived from CTRW (β, α, and D_m) and diffusion-weighted imaging (DWI) were calculated and compared between higher- and lower-proliferative groups. Moreover, logistic regression (LR) analysis was utilized to find independent predictive factors and to develop a combined diagnostic model. Diagnostic efficacy was assessed utilizing the DeLong test, AUC, and calibration curves.

The μ value was substantially greater in the high-proliferative group than in the low-proliferative group, whereas α, β, D_m, and ADC values were notably lower in the high-proliferative group. Moreover, the respective AUCs were 0.875 (95% CI:0.785 ~ 0.937), 0.860 (95% CI:0.768 ~ 0.926), 0.815 (95% CI:0.716 ~ 0.891), 0.789 (95% CI:0.687 ~ 0.870), and 0.769 (95% CI:0.668 ~ 0.854). LR analysis confirmed μ, α, D_m, and ADC as independent predictors of Ki-67 status in CC. The combined diagnostic model incorporating these predictors achieved the highest performance (AUC = 0.970, 95% CI:0.908 ~ 0.995), significantly outperforming individual modalities such as vMRE (μ), DWI (ADC), and single parameters α, β, and D_m (Z = 2.891, 4.271, 2.692, 3.102, and 4.064, respectively; all P < 0.05). Furthermore, calibration curves demonstrated favorable stability for the combined diagnostic model.

Both vMRE and CTRW show better performance than conventional DWI in differentiating high- from low-proliferative CC. The combined use of μ, α, Dm, and ADC provides preliminary, hypothesis-generating evidence as a potential imaging biomarker for evaluating Ki-67 status, which requires further validation in larger independent cohorts.