LG AI Research’s “Exaone Path 2.5,” a disease and genetic analysis artificial intelligence model, has demonstrated superior performance in cancer diagnosis compared to leading global AI models, including those developed by Harvard University and Microsoft, according to a report released Tuesday.
The AI model achieved an accuracy rate of 76.75% in a recent evaluation conducted by LG AI Research, surpassing the performance of models such as Harvard Medical School Professor Mahmoud’s team’s TITAN (73.20%) and UNI2-h (76.16%), Microsoft’s Gigapath (71.43%), and Bio-optimus’s H-optimus-0 (75.78%). The evaluation utilized clinical data from hospitals in both Korea and the United States, focusing on the detection of tumors and genetic mutations in colorectal and non-small cell lung cancers.
LG AI Research compared Exaone Path 2.5 with other open-source medical AI models, highlighting its unique ability to learn even when pathology images and multi-omics (genetic information) are not directly linked. Traditionally, AI requires matched pathology images and genetic data for accurate diagnosis. Exaone Path 2.5 can independently identify correlations, maximizing learning efficiency and achieving high accuracy even with a smaller model size, according to LG officials.
Further validating its capabilities, Exaone Path 2.5 achieved 69.8% accuracy when assessed using the performance metrics developed by Mahmoud’s research team, placing it second to TITAN’s 71.4%.
The findings, reported by multiple Korean news outlets including Nate News and Daily Medical, underscore a growing trend of Korean AI technology gaining prominence on the global stage. The development of Exaone Path 2.5 is expected to accelerate the application of AI in clinical settings, potentially improving diagnostic accuracy, and efficiency.
