Abstract 1239P
Background
The five-year overall survival for oesophageal adenocarcinoma remains poor at approximately 15%-25%. This highlights the need for improved therapeutic approaches. Radiomics studies combine quantitative medical image features with genomics to assess tumour phenotype as a “virtual biopsy.” These have shown promising results defining medical image features that predict treatment response and survival in oesophageal squamous cell carcinoma. Currently, there are very few published studies examining radiomics profiles in OAC. A unique opportunity exists in further exploration of radiomics features and how they can be used for prognosis.
Methods
PET/CT image analysis was performed on pre-treatment scans from 50 oesophageal adenocarcinoma patients who underwent standard of care treatment. A filtration-histogram method was used to perform the CT texture analysis (CTTA) and evaluate the prognostic value of imaging biomarkers. RNA sequencing (RNAseq) and immunohistochemistry characterised the molecular pathways associated with these biomarkers. A deep learning workflow was used to validated biomarkers identified using the filtration-histogram method.
Results
Survival analysis determined three PET/CT image biomarkers were significantly associated with both disease specific survival and progression-free survival in these patients. In a validation cohort, a deep learning workflow confirmed the clinical significance of these image biomarkers. RNAseq was performed on a subset of patients to identify the gene pathways associated with these biomarkers. RNAseq confirmed that lower CD8 T cell expression in the tumour correlated with the poor survival group predicted using CT markers. Immunohistochemistry confirmed markers identified by RNAseq.
Conclusions
PET/CT imaging is an essential component of cancer diagnostics and has an established role in patient management. This study identifies PET/CT image features representing immune phenotypes and demonstrates their value as prognostic biomarkers.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The University of Queensland.
Funding
National Health and Medical Research Council of Australia, PA Research Foundation.
Disclosure
All authors have declared no conflicts of interest.