Advancing cancer detection and treatment using longitudinal routine clinical data

Oncoformer, a multimodal transformer trained on routine laboratory tests and chest X-rays from over 3.6 million individuals, detects cancer, infers tumor stage, and stratifies both treatment response and recurrence risk using data already collected in everyday clinical care. Published in Cell, the model operates across multiple cancer types without requiring additional specialized testing. The scale of validation and the breadth of clinical endpoints — from early detection through post-treatment surveillance — position Oncoformer as a step toward risk-adapted oncology built entirely on existing health-system data.