Multimodal AI Model Oncoformer Enhances Pan-Cancer Detection Using EHR and Chest X-Ray

Oncoformer, a multimodal transformer trained on 3.67M individuals, achieved a pan-cancer diagnosis AUROC of 0.956 and predicted future cancer up to 1 year before diagnosis with AUROC 0.869, using longitudinal EHR and chest X-ray data.

• Validated on independent external cohorts including UK Biobank.
• Tumor stage inference achieved mean AUROC >0.90, independently validated against postoperative pathology.
• Model demonstrated significant recurrence-free survival stratification across ten cancer types (all p < 0.01).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.