Identification of Lung Adenocarcinoma Subtypes using Radiogenomics and Deep Learning

Identifying lung adenocarcinoma subtypes is advantageous not only for individual patient care but also for improved patient selection for current treatment options. However, cancer subtyping using advanced image traits and somatic mutations is still in the premature stage mainly due to tumor heterogeneity, natural diversity, and complex genotype-environment interactions. Our goal has been to enable precise lung adenocarcinoma subtyping by developing a unified radiogenomic framework using deep learning. Our framework has two specific aims: 1) to standardize CT images to facilitate deep image trait extraction and 2) to identify significant image trait-somatic mutation associations for lung adenocarcinoma subtyping. Taking advantage of recent advances in high-resolution CT machines, next-gen sequencing, and deep learning, we expect to identify key characteristics towards the development of new lung adenocarcinoma subtyping criteria, which once developed, could remarkably change treatment paradigms and significantly improve prognosis.