38 Modelling Class Dependencies for Lung Cancer Subtyping from Digitised Pathology Images

Lung cancer's diverse histological subtypes require precise identification for effective treatment. Current computational pathology methods lack specificity and require resource-intensive annotations. This study addresses these challenges by constructing a new multi-label dataset where cancer subtypes and associated patterns are both annotated on slide level. A novel computational model is introduced for linking the cancer subtypes and histological patterns in a weakly-supervised fashion.