Patients with presumed early cervical cancer being considered for curative surgery

Pixel by pixel analysis of ADC from tumour regions of interest will be used to derive histograms and the 10th, 25th, 50th, 75th and 90th centile values and histogram skew, kurtosis and entropy values for each tumour will be recorded. The number of regions will vary with tumour volume in each patient, given the mean size of lesions in patients studied in a pilot set, it is envisaged that we will analyse 1-5 regions of interest in each tumour. The data will be analysed separately by 2 observers who will be blinded to the histopathological characteristics and assuming good inter-observer agreement, an average of their results will be used for data analysis. The difference in mean centile ADC values of the distribution and the skewness, kurtosis and entropy values between squamous vs. adenocarcinoma, well/moderately differentiated vs. poorly differentiated tumour regions and the absence vs. the presence of lymphovascular space invasion in that region will be determined.

The centile measures of the ADC values will be summarised using means and standard deviations if the data are normally distributed and medians and ranges if the data are non-normally distributed. The unpaired t-test will be used to analyse normally distributed data and the Mann Whitney U test for non-parametric data. Point estimates for the differences between centile ADC values and other summary measures for tumour type, grade and presence or absence of lymphovascular space invasion will be derived and 95% confidence intervals attached where appropriate. The Bonferroni correction method will be used to adjust significance levels for multiple comparisons.

The key clinical question is whether or not the imaging can differentiate poor prognosis tumours from good prognosis ones from the histogram metrics.