AI and imaging-based cancer screening: getting ready for prime time

Pancreatic cancer is a major cause of cancer-associated mortality, with a dismal overall prognosis that has remained virtually unchanged for many decades. Currently, prevention or early diagnosis at a curable stage is exceedingly difficult; patients rarely exhibit symptoms and tumours do not display sensitive and specific markers to aid detection. Pancreatic cancers also have few prevalent genetic mutations; the most commonly mutated genes are KRAS, CDKN2A (encoding p16), TP53 and SMAD4 - none of which are currently druggable. Indeed, therapeutic options are limited and progress in drug development is impeded because most pancreatic cancers are complex at the genomic, epigenetic and metabolic levels, with multiple activated pathways and crosstalk evident. Furthermore, the multilayered interplay between neoplastic and stromal cells in the tumour microenvironment challenges medical treatment. Fewer than 20% of patients have surgically resectable disease; however, neoadjuvant therapies might shift tumours towards resectability. Although newer drug combinations and multimodal regimens in this setting, as well as the adjuvant setting, appreciably extend survival, ∼80% of patients will relapse after surgery and ultimately die of their disease. Thus, consideration of quality of life and overall survival is important. In this Primer, we summarize the current understanding of the salient pathophysiological, molecular, translational and clinical aspects of this disease. In addition, we present an outline of potential future directions for pancreatic cancer research and patient management.

Medical artificial intelligence (AI) offers great potential for recognizing signs of health conditions in retinal images and expediting the diagnosis of eye diseases and systemic disorders 1 . However, the development of AI models requires substantial annotation and models are usually task-specific with limited generalizability to different clinical applications 2 . Here, we present RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled retinal images and provides a basis for label-efficient model adaptation in several applications. Specifically, RETFound is trained on 1.6 million unlabelled retinal images by means of self-supervised learning and then adapted to disease detection tasks with explicit labels. We show that adapted RETFound consistently outperforms several comparison models in the diagnosis and prognosis of sight-threatening eye diseases, as well as incident prediction of complex systemic disorders such as heart failure and myocardial infarction with fewer labelled data. RETFound provides a generalizable solution to improve model performance and alleviate the annotation workload of experts to enable broad clinical AI applications from retinal imaging. RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled images, is trained on 1.6 million unlabelled images by self-supervised learning and then adapted to disease detection tasks with explicit labels.