Genomic data, histology, radiology and clinical features all have independent relevance for clinical decision-making and are thus collected throughout the diagnosis and treatment of cancer. Despite the ubiquitous clinical usage of these data, recent studies have demonstrated that the implementation of deep-learning methods can reveal additional prognostic information encoded in these modalities. More specifically, deep learning has been used to identify patients with distinct genomic and clinical features based on hematoxylin and eosin (H&E)-stained slides, and computed tomography (CT) imaging alone. What were once thought to be non-intersectional data modalities with a narrow purview are now understood to be different lenses for viewing a patient’s cancer, highlighting overlapping facets of the disease based on the data type. Although previous work has demonstrated some degree of synergy in the information from genomic, histological and radiological data, the degree to which each data type offers unique, clinically relevant information remains unknown.
展开▼
