Cervical cancer represents the fourth leading cause of cancer death in women, with 311,000 deaths in 2018 worldwide. Treatment depends mainly on the stage at diagnosis, as assessed by the International Federation of Gynecology and Obstetrics (FIGO) staging system.
Although survival rates for women with locally advanced cervical cancer (LACC) are improving, there is still a high rate of persistence of residual tumour after chemotherapy, varying between 35%–63%. Literature data support the advantage of neoadjuvant chemotherapy (NACRT) followed by radical hysterectomy for the removal of potential radio-resistant and chemo-resistant neoplastic foci, improving local control. As a matter of fact, pathological complete response (pCR) has been associated with higher disease-free and long-term survival.
A personalised approach based on pre-treatment prediction of pCR may therefore have a significant impact on the management of patients affected by LACC.
Radiomics is an expanding field of clinical research that provides the possibility to analyse and quantify intra-tumoural heterogeneity in a non-invasive way, offering the chance to individuate a single patient’s risk group and allowing treatment to be customised according to the predicted outcome.
As far as our research group is aware, no studies have yet created a radiomics model for pCR prediction in locally advanced cervical cancer.
Our aim is to create a radiomics model that is able to predict pCR after NACRT in cervical cancer by analysing MRI data acquired before starting the treatment, in order to offer the chance to modulate treatment on new patients, in the frame of fully personalised clinical management. The model was initially developed by analysing patients coming from an internal training set. Then, the model was applied to an external validation set of patients, with successful results.
We included 183 patients in the analysis, coming from two different institutions, with the same treatment protocol (NACRT plus radical hysterectomy). In both centres, patients were scanned using 1.5T MR machines. The gross tumour volume was contoured by two radiologists who are experts in gynaecological imaging, in consensus, using a radiotherapy treatment planning system. More than 1,000 radiomic features belonging to four families (fractal, statistical, textural and morphological features) were extracted from the images. The feature selection and model training were carried out following an iterative method developed for this. At the end of the elaboration process, 15 models were obtained and the area under the curve for all the models was taken into account.
This is the first radiomics-based pCR prediction model on pre-treatment staging MRI in patients affected by locally advanced cervical cancer undergoing NACRT. The choice of pCR as an outcome parameter represents the strength of this study; in fact, surgery after chemotherapy offers the possibility of achieving a certain pathological diagnosis.
The proposed model allows the probability of having a pCR to be predicted using pre-treatment imaging, so it would be possible for clinicians to establish whether a patient will be a good responder or not before treatment starts. In this way, clinicians have the possibility to design new protocols of dose escalation or dose deintensification on the basis of a single patient’s predicted outcome.
Luca Russo, MD is a radiology resident at the Institute of Radiology, Fondazione Policlinico Universitario A. Gemelli IRCCS – Università Cattolica del Sacro Cuore in Rome, Italy.
Research Presentation Session
RPS 1007 Imaging in pregnancy and female tumours
Pre-treatment MRI radiomics-based response prediction model in locally advanced cervical cancer
L. Russo, B. Gui, S. Persiani, M. Miccò, L. Boldrini, D. Cusumano, R. Autorino, G. Ferrandina, R. Manfredi; Rome/IT
Read the full abstract in the ECR 2020 Book of Abstracts
Russo L, et al. (2020) Pre-treatment MRI radiomics-based response prediction model in locally advanced cervical cancer. Abstract RPS 1007-7 in: ECR 2020 Book of Abstracts. Insights Imaging 11, 34 (2020). DOI 10.1186/s13244-020-00851-0