Optimisation of a diagnostic model to predict the effectiveness of chemoradiotherapy for cervical cancer in a group of patients with comorbid conditions: cohort single-center retrospective study

Abstract

INTRODUCTION: Radiomics is a promising area of diagnostics. In clinical practice, ultrasound and magnetic resonance imaging are widely used for Cervical Cancer (CC). The lack of standards when carrying out examinations entails the problem of distinguishing different signs, i.e. there is no possibility to compare results of different institutions.OBJECTIVE: To review radiological diagnostic procedures and optimize a model to enable expanded large-scale multicentre mathematical analysis of radiological findings in comorbid women with CC.MATERIALS AND METHODS: The data from 362 magnetic resonance imaging (MRI) procedures (Philips Achieva, The Netherlands, 1.5T), 500 pelvic ultrasound procedures (US), and 500 retroperitoneal US in 77 comorbid women with cervical squamous cell cancer and cardiovascular disease, carried out between 2012 and 2022, were retrospectively examined. FIGO pretreatment stage 1А–4А. Age: 48.3±13.1. Follow-up period: 3.7±1.3 years.Statistics: Data analysis was carried out using the Stata 13 program (StataCorpLP, CollegeStation, TX, USA). The normality of the distribution of features was assessed using the Shapiro-Wilk criterion. The condition of equality of variances of the distribution of features was calculated according to the Leven criterion. For descriptive statistics of normally distributed features with equality of variances, the calculation of averages and standard deviations was used. Qualitative variables are represented as numbers (%). Logistic regression is performed. The significance level for all the methods used is set as p<0.05.RESULTS: The possibility of segmentation was 2.6% according to US and 100% according to MRI. We analyzed 1443 T2 TSE, 531 T1 TSE, 563 diffusion-weighted images (DWI), 389 STIR, 1987 post-contrast series (in 272 cases (75%) the study was accompanied by contrast agent administration). An MRI model for subsequent feature extraction in patients with CC should consist of T2TSE in the sagittal plane, DWI in the axial plane with automatic construction of apparent diffusion coefficient (ADC) maps.The most reproducible and valuable components of the model are found to be the DWI with automatic ADC map. The ADC value from the parametral fat significantly increased the probability of recurrence, and the cut-off point for ROC analysis was 1.1×10–3 mm2/sec.DISCUSSION: An analysis of medical ultrasound and MRI images in terms of their value for radiomics was carried out. According to the results, MRI is the preferred method. An important next step is to standardize series to extract additional value from diagnostic studies and to carry out multicentre retrospective studies using a multicomponent model.CONCLUSIONS: MRI is a reproducible and frequently used method with the ability to extract additional value from images. T2 TSE in the sagittal plane and DWI in the axial plane with automatic ADC map, followed by segmentation of the parametral area adjacent to the tumor, are considered the most frequently used techniques. Postcontrast imaging are not a reproducible technique and have no added value. A model MRI procedure to determine additional textural characteristics in patients with СС consists of T2-TSE in the sagittal plane, DWI in the axial plane with automatic ADC map.