Can Artificial Intelligence applied to a pre-treatment 18F-FDG PET/CT scan be used to predict two year disease recurrence for patients wIth Oesophageal cancer?


We propose investigating the 2 year disease recurrence-free survival for oesophageal cancer patients (Foley, et al., 2018; Xiong, et al., 2018), based on the pre-treatment PET scan (Sah, et al., 2019; Tixier, et al., 2011) and will compare both a classical approach using radiomics to inform a random forest machine learning algorithm (Paul, et al., 2017; Larue, et al., 2018) and a CNN approach using the architecture proposed by Ypsilantis et al (2015).

One issue for our dataset is that a high quality 3D PET/CT scanner was used for the vast majority of patients from mid-2016. Previously a 2D scanner was used for the majority of patients; therefore, we propose including only patients that were scanned on a 3D scanner.

Furthermore, we propose correlating with 2 year disease recurrence-free survival (rather than 3 year survival (Larue, et al., 2018)) such to include the maximum number of patients into the study. To date, this particular combination of comparing two optimal machine learning methodologies with the 2 year disease recurrence-free survival has not been investigated for Oesophageal cancer patients, using pre-treatment PET imaging.