PathoVision AI Pathology Diagnostic Accuracy Surpasses Humans: Digital Pathology Enters AI-Led Era

Pathological diagnosis is the gold standard for cancer confirmation, but the world faces a severe pathologist shortage. The World Association of Pathology data shows there are only about 120,000 practicing pathologists globally, while over 300 million cases require pathological diagnosis annually. In sub-Saharan Africa, there is approximately 1 pathologist per 5 million population.

On July 1, 2028, Google Health AI published a paper in the New England Journal of Medicine reporting that its PathoVision AI pathology diagnostic system achieved a diagnostic accuracy of 97.3% in a large-scale clinical validation covering 12 common cancers, surpassing the 200 senior pathologists in the control experiment (average accuracy 94.1%). More importantly, PathoVision's false negative rate (missed diagnosis rate) was only 1.2%, compared to 3.8% for human pathologists.

PathoVision's architecture is based on an improved Vision Transformer, trained on over 50 million annotated pathology slides. The system can process digitized pathology slides from different staining methods and generate comprehensive reports including malignancy probability, grading scores, and molecular marker predictions.

Google Health AI lead Dr. Greg Corrado said: "PathoVision isn't replacing pathologists—it's freeing them from repetitive screening to focus on complex and rare cases. In a world with severe pathologist shortages, AI-assisted diagnosis is key to expanding healthcare access."

The technology has received EU CE certification and is seeking FDA approval. In China, Tencent Miying has reached a technology cooperation agreement with the PathoVision team to launch a joint version for the Chinese market.

Challenges include digitization of pathology slides—most hospital pathology labs worldwide still use traditional optical microscopes—and legal liability questions about who has the final say when AI and human pathologists disagree.