Lunit unveils AI research to participate in AstraZeneca and Daiichi Sankyo at ESMO
The two companies also announced results from follow-up research on lung cancer biomarkers and analysis of the environment around pancreatic cancer cells.
Lunit will present research results from its breast cancer AI analysis project, which involves global pharmaceutical companies AstraZeneca and Daiichi Sankyo. Researchers will also publish a study comparing AI results with pathologists' decisions about lung cancer biomarkers and another study examining the environment around cancer cells in patients with pancreatic cancer.
On the 20th, Lunit announced that it will present five research summaries using its AI biomarker platform, Lunit Scope, at the 2026 European Society for Medical Oncology (ESMO) meeting in Madrid, Spain, from the 23rd to the 27th.
The main presentation will share the first research results from AIforHER, a global project using AI to analyze HER2 in breast cancer. The project is led by ICO, a French cancer organization, and includes pharmaceutical companies and medical institutions such as AstraZeneca and Daiichi Sankyo. It aims to develop and test AI programs that can help tailor breast cancer treatment to individual patients.
The project team compared AI tools from several companies for patients with very low levels of HER2, a protein that helps doctors decide which breast cancer treatments to use. It has become increasingly important to identify even small differences in HER2 levels because patients whose tumors have "HER2 ultra-low expression" may also receive antibody-drug conjugate (ADC) treatment. After this comparison, Lunit was selected as one of only two companies to take part in a follow-up study. Its AI tool is now being tested in real clinical settings.
In the lung cancer study, researchers analyzed tissue samples taken during surgery from 993 patients with non-small cell lung cancer who were enrolled in a research project led by Japan's National Cancer Center. Pathologists and Lunit Scope PD-L1 separately assessed the amount of PD-L1 protein in the samples. PD-L1 is a marker used to help guide treatment with cancer immunotherapy drugs. The two assessments agreed in 73% of cases.
The assessments differed more often when PD-L1 levels were low. Pathologists judged that PD-L1 levels were below 1%, but the AI tool judged them to be above 1% in 211 cases. In 204 of those cases (97%), the AI estimated that PD-L1 levels were between 1% and 10%. The researchers said further clinical studies are needed to find out how these differences in judgment affect actual treatment results.
Researchers also studied 181 patients with pancreatic ductal adenocarcinoma, a type of pancreatic cancer. They examined how outcomes varied based on the tissue material surrounding the cancer cells and on genetic mutations. Using Lunit Scope IO to analyze tissue images, they found that, across the overall patient group, those with more of this surrounding tissue material had less time before their cancer worsened or they died. However, among patients with a specific KRAS gene mutation called G12D, the time was longer for those with more of this material. Median overall survival was 43.8 months for patients without a KRAS mutation, 19.5 months for patients with other KRAS mutations, and 15.1 months for patients with the G12D mutation.
"At this conference, we will present important studies examining how AI can help in areas that are difficult to assess through microscope examination alone, such as HER2 ultra-low expression," said Seo Beom-seok, CEO of Lunit. "We will continue to expand our partnerships with global pharmaceutical companies and research institutions so that Lunit Scope can provide more precise ways to classify cancer patients."