The use of machine learning/deep learning in PET/CT interpretation to aid in outcome prediction in lymphoma

Lymphoma is a haematopoietic malignancy consisting of two broad categories: Hodgkin lymphoma (HL) and non-Hodgkin lymphoma (NHL). These categories can be further split into subtypes with classical HL (cHL) and diffuse large B cell lymphoma (DLBCL) being the commonest subtypes. The gold standard imaging modality for staging and response assessment for cHL and DLBCL is 2-deoxy-2-[fluorine-18]fluoro-D-glucose (FDG) positron emission tomography/computed tomography (PET/CT), with patients having a worse prognosis if they do not demonstrate complete metabolic response (CMR). However, approximately 15% of patients will relapse even after CMR. Therefore, being able to identify patients who are likely to relapse it may be possible to stratify treatment early to improve patient outcomes. The aim of this project is to develop and test image derived predictive models based on the baseline PET/CT to risk stratify patients pre-treatment.