Using imaging mass cytometry, the tumour and immunological spatial landscapes of 416 lung adenocarcinomas are characterized, which, when combined with deep learning, can predict clinical outcomes with high accuracy.