LG launches upgraded pathology AI model

A view of LG Sciencepark in Seoul, home to the headquarters for LG AI Research / Courtesy of LG Group

Hwang Tae-hyun, professor at Vanderbilt University Medical Center / Courtesy of LG Group

LG AI Research unveiled its next-generation medicine artificial intelligence (AI) model — the EXAONE Path 2.0 — as part of its push to revolutionize cancer diagnosis and drug development Wednesday.

The AI research arm of LG Group launched the EXAONE Path 1.0 in August last year, which the company said represented a “significant advancement” in the use of multimodal AI in oncology.

The new EXAONE Path 2.0 has been trained on high-quality data to analyze genetic mutations, expression patterns and subtle structural changes in human cells and tissues through pathology slide images, the AI lab said.

This enables earlier cancer diagnosis and more accurate prognosis, thereby helping doctors set personalized treatment plans. It is also expected to be used for more efficient drug development.

Leveraging multimodal data — including DNA, RNA and pathology images — the EXAONE Path 2.0 integrates high-resolution whole slide images (WSI), each comprising gigabytes of cellular data with multiomics information.

Such images are traditionally broken into thousands of patches for analysis, which can cause a phenomenon known as feature collapse, when AI focuses too narrowly on individual cells and loses track of the overall context.

To address the challenges, LG AI Research developed a new training method that incorporates both the full WSI and individual patches. As a result, the EXAONE Path 2.0 achieved a gene mutation prediction accuracy rate of 78.4 percent, significantly improving diagnostic reliability without the need for costly genetic sequencing.

“The EXAONE Path 2.0 reduces the typical two-week turnaround for genetic testing to under a minute,” said Park Yong-min, head of AI business at LG AI Research.

“This can be crucial for cancer patients, helping doctors and pharmaceutical firms rapidly identify genetic mutations and match them with targeted therapies.”

LG AI Research expects the EXAONE Path 2.0 to play a crucial role in real-time clinical decision-making, including identifying novel biomarkers, predicting patient responses and enabling real-world evidence studies in clinical trials.

To accelerate the drive, LG AI Research is partnering with a medical research team at Vanderbilt University Medical Center to codevelop a multimodal medical AI platform.

The joint project will use real patient samples and treatment data from clinical trials to refine AI capabilities in diagnosing diseases at earlier stages and developing individual treatment strategies based on genetic profiles, according to LG AI Research.

“We are creating an AI platform that can be actively used by clinicians in real-world settings, and even transform how new drugs are developed,” said Hwang Tae-hyun, a professor at Vanderbilt University Medical Center.