Development and evaluation of a cervical cancer-specific restriction spectrum imaging (RSI) model

Cervical cancer is the fourth most common cause of cancer and cancer related mortality among women worldwide. MRI is standard of care for cervical cancer staging. The objective of this project is to evaluate the feasibility and performance of an advanced diffusion-weighted imaging (DWI) MRI technique for cervical cancer.

Forty-four patients with cervical cancer and 22 healthy volunteers underwent 3 T pelvic MRI with reduced field-of-view (FOV) multi-shell diffusion-weighted imaging (DWI; b = 0-3000 s/mm²). Patients were divided into model development and independent testing cohorts. Healthy volunteers provided reference distributions for Z-score mapping. RSI models with multi-exponential components were fitted to tumor signals. For each model, the outputs were voxel-wise compartmental signal contributions (C_i,N) from each water component within tissue. Model performance was assessed using the Bayesian Information Criterion (BIC), tumor contrast-to-noise ratio (CNR), and Z-scores. Conventional apparent diffusion coefficient (ADC) was estimated from full-FOV DWI (b = 50-1000 s/mm²).

The tetra-exponential model achieved the highest CNR, with only a 6% BIC increase relative to the bi-exponential model. In an independent cohort, CNR was significantly higher in restricted compartments C_1,3 (median 14.3) and C_1,4 (median 26.2) compared with ADC. Tumor Z-scores were elevated in C_1,3 (median 3.8) and C_1,4 (median 8.0), whereas other compartments remained near zero.

RSI with Z-score mapping improved cervical cancer conspicuity compared with ADC, producing interpretable maps that localized tumor signal to restricted diffusion compartments. Based on BIC and CNR, the tetra-exponential RSI model was optimal, supporting RSI as a feasible non-contrast imaging strategy for cervical cancer evaluation.