Advanced MRI for developing more personalized treatment strategies in uterine cervical cancer

Cervical cancer is a major threat to global health, ranking fourth worldwide in terms of incidence and as a cause of cancer-related mortality among women. While early-stage disease has excellent prognosis, the survival rates for locally advanced or metastatic disease are markedly lower. Key clinical challenges in managing cervical cancer include accurately identifying high-risk patients and effectively monitoring treatment responses for optimal tailoring of therapeutic approaches and surveillance protocols. 

The overall objective of this thesis was to identify reliable and novel magnetic resonance imaging (MRI)-based biomarkers relevant for prognosis and treatment in cervical cancer, aiming to refine risk stratification and promote personalized treatment strategies that improve patient outcomes.

Material and methods: Papers I–IV present retrospective analyzes of MRI-derived imaging data from cervical cancer patients treated at Haukeland University Hospital from 2002 to 2022. Three radiologists blinded to clinical information independently read and systematically recorded findings from all pretreatment MRIs. 

Papers I–II studied a cohort of 416 cervical cancer patients who had pretreatment routinely performed pelvic MRI (2002–2017). Paper I examined the interobserver agreement (evaluated using kappa coefficients [κ]) for pretherapeutic MRI-based 2018 International Federation of Gynecology and Obstetrics (FIGO) staging parameters, and assessed their prognostic value in comparison to other clinicopathological markers. Paper II explored the value of various pretreatment MRI-based tumor size measurements for predicting disease-specific survival (DSS) and evaluated their interobserver reproducibility (as measured by intraclass correlation coefficients [ICCs]).

In Paper III, 133 cervical cancer (2018 FIGO stage ≥IB1) patients having an MRI protocol encompassing T2-weigted imaging (T2WI) and diffusion-weighted imaging (DWI) were included (2009–2017). The patients were allocated into training (cohort_T: n=89) and validation (cohort_V: n=44) cohorts. Radiomic features from manually segmented primary tumors were extracted from T2WI and DWI series, and radiomic signatures from T2WI only (T2_rad) and combined T2WI and DWI (T2+DWI_rad) for DSS prediction were developed and validated. The prognostic performance of these signatures was evaluated and compared with MRI-derived maximum tumor size (MAX_size) of ≤/>4 cm and 2018 FIGO stages I–II/III–IV. 

Paper IV included 110 patients with locally advanced cervical cancer undergoing concurrent chemoradiotherapy (CCRT) (2007–2022) and sequential MRI. These patients underwent routine MRI at both baseline and midtreatment (after ~4 weeks of CCRT), displaying visible tumors on both scans. Radiomic features from T2WI were extracted from manually segmented tumors at both time points, and temporal changes (delta features) were calculated. Two radiomic signatures for prediction of progression-free survival (PFS), Delta_rad (based on delta features) and Pre-CCRT_rad (based on pretreatment features), were developed in a training cohort (cohort_T: n=73) and validated in a validation cohort (cohort_T: n=37). The signatures were evaluated against 2018 FIGO stages I–IV and baseline MRI-derived maximum tumor diameter (Tumor_max; ≤2 cm; >2 and ≤4 cm; >4 cm). The study also investigated the association between changes in T2WI radiomic features during CCRT and PFS.

Results: In Paper I, the interobserver agreement was substantial for the MRI-derived staging parameters: tumor size >2 cm, tumor size >4 cm, tumor size categories (≤2 cm; >2 and ≤4 cm; >4 cm), parametrial invasion, vaginal invasion, and enlarged lymph nodes (κ values in the range of 0.61–0.80). An increase in MRI-based tumor size category (≤2 cm; >2 and ≤4 cm; >4 cm) was associated with a stepwise decline in survival (p≤0.001 for all). Tumor size >4 cm and parametrial invasion identified on MRI were linked to aggressive clinicopathological features, and the incorporation of these MRI-based staging parameters improved risk stratification compared to relying solely on clinical assessments.

In Paper II, MRI-based measurements of the largest tumor diameters in three orthogonal planes and the maximum diameter irrespective of plane demonstrated excellent interobserver reproducibility (ICCs of 0.83–0.85). All MRI tumor size variables (cm) yielded high area under the time-dependent receiver operating characteristic (tdROC) curves (AUC) for 5-year DSS prediction (AUCs of 0.81–0.84) and had a significant impact on outcome (hazard ratios [HRs] of 1.42–1.76, p<0.001 for all). However, only maximum diameter irrespective of plane was an independent predictor of survival (HR of 1.51, p=0.03), with an optimal cutoff of ≥4 cm. 

In Paper III, the radiomic signatures T2_rad and T2+DWI_rad yielded tdROC AUCs for predicting 5-year DSS in cohort_T/cohort_V (AUC_T/AUC_V) of 0.80/0.62 and 0.81/0.75, respectively. Both signatures demonstrated prognostic performance metrics comparable to or better than that of MAX_size (0.69/0.65) and FIGO (0.77/0.64). After adjusting for FIGO, both signatures were significant predictors of survival (HR_T/HR_V for T2_rad: 4.0/2.5 and T2+DWI_rad: 4.8/2.1, p≤0.04 for all). Incorporating T2_rad or T2+DWI_rad to FIGO significantly enhanced survival prediction compared with FIGO alone in cohort_T (AUC_T of 0.86 and 0.88 vs. 0.77 for FIGO alone, p≤0.008 for both combined models). A similar trend was noted in cohort_V for FIGO with T2+DWI_rad (AUC_V of 0.75 vs. 0.64 for FIGO alone, p=0.07). A high radiomic score for T2+DWI_rad was strongly linked to reduced DSS in both cohorts (p≤0.01 for both). 

In Paper IV, Delta_rad (AUC_T/AUC_V: 0.74/0.79) slightly outperformed Pre-CCRT_rad (0.72/0.75) in predicting 5-year PFS, and both signatures clearly surpassed that of FIGO (0.61/0.61) and Tumor_max (0.58/0.65). In total, four radiomic features within Delta_rad and Pre-CCRT_rad showed significant differences in delta feature values between progressors and non-progressors, with consistently lower values in progressors (p≤0.03 for all). High scores for Delta_rad and Pre-CCRT_rad were associated with poor PFS (p≤0.04 for Delta_rad in cohort_T /Pre-CCRT_rad in both cohorts; p=0.11 for Delta_rad in cohort_V).

Conclusions: The interobserver agreement for central MRI-derived 2018 FIGO staging parameters in cervical cancer was substantial. MRI improved the identification of patients with aggressive clinicopathological features and poor survival, underscoring its potential to refine prognostication and enable tailored treatment for cervical cancer patients (Paper I). Among various MRI-derived tumor size measurements in cervical cancer, maximum diameter irrespective of plane was the only independent predictor of DSS, with ≥4 cm as the optimal cutoff value. Interobserver reproducibility for all the MRI tumor size measurements was excellent (Paper II). Radiomic signatures derived from T2WI and T2WI with DWI may offer additional value for pretreatment risk assessment and for guiding personalized treatment strategies in cervical cancer (Paper III). Delta and pretreatment radiomic signatures from T2WI equally enhance early prognostic assessments in locally advanced cervical cancer, surpassing the predictive capabilities of 2018 FIGO stage and MRI-assessed maximum tumor diameter (Paper IV).