Artificial intelligence (AI) is increasingly seen as a valuable assistant in pathology due to its ability to recognize patterns in tissue samples, according to Andrew Norgan from the Mayo Clinic. While AI has excelled in visual analysis since its inception, current models remain experimental. A recent attempt, called Atlas, developed by Aignostics and the Mayo Clinic, was trained on 1.2 million samples and tested against six leading AI pathology models. Atlas outperformed competitors in six out of nine tasks, achieving a 97.1% agreement rate with human pathologists for colorectal cancer classification. However, it still showed limitations, such as a 70.5% accuracy rate for prostate cancer biopsies. Experts suggest that AI models need to exceed 90% accuracy for clinical use, yet even less-than-perfect models could enhance pathologists’ efficiency. A significant barrier to progression is the lack of digitized pathology data, with fewer than 10% of U.S. practices fully digitized, hindering AI training efforts.
Source link
