Introduction/Background Pelvic magnetic resonance imaging (MRI) is an important part of primary diagnostic workup in cervical cancer (CC), with MRI-assessed tumour size and pelvic tumour extent routinely guiding treatment decisions. Extraction of MRI-derived radiomic tumour features could improve cancer prognostics and may also reveal novel targets for treatment.

Methodology We manually segmented 3D volumes in 132 primary CC tumours and extracted 293 whole-volume radiomic MRI features. Unsupervised hierarchical clustering yielded three distinct patient clusters (Cluster 1 [n=52]; 2 [n=46]; and 3 [n=34]). Overlapping clinicopathologic, genomic (whole exome sequencing, n=65), transcriptomic (L1000 arrays, n=73) and molecular biomarker (n=84) data were utilized to characterize each cluster.

Results Patients in Cluster 2 and 3 had significantly reduced disease-specific survival (DSS) (hazard rate [HR]: 3.33; p=0.008) compared with patients in Clusters 1, even after adjusting for International Federation of Gynecology and Obstetrics (FIGO) 2018 stage and age (adjusted HR: 2.51; p=0.045). Cluster 3 tumours associate with high stage (p<0.001), large tumours (p<0.001), squamous histology (p=0.015), p53 negative or -overexpressing tumours (p=0.04) and aberrant TP53, -MYC and -MTORC1 signalling. The intermediate-risk Cluster 2 associates with increased and aberrant cell cycle- and Hippo signalling, suggesting CDK2/4 and YAP-TEAD inhibitors as plausible treatment options. The low-risk Cluster 1 associates with increased immune cell signalling.

Conclusion This study links radiomic signatures to distinct genomic profiles that may potentially aid in prognostication and tailoring of treatments and follow-up plans for cervical cancer patients.

Disclosures The authors declare no conflict interests.