Lunit announced the presentation of seven studies at the American Society of Clinical Oncology (ASCO) 2024 Annual Meeting in Chicago, from May 31 to June 4. Lunit will present detailed findings on several innovative studies, including the identification of HER2 ultra-low expression in breast cancer using AI-based quantification, and a deep learning-based model integrating chest CT and histopathology analysis for predicting immunotherapy response in non-small cell lung cancer (NSCLC). In a poster presentation, Lunit's AI-powered HER2 analyzer, Lunit SCOPE HER2, demonstrated the ability to identify HER2 ultra-low expression and differentiate it from true HER2-negative cases in breast cancer patients using continuous subcellular quantification from HER2 immunohistochemistry (IHC) images. According to findings presented at ASCO 2022, HER2-targeted antibody-drug conjugates (ADCs) can effectively target tumor cells even in HER2-low breast cancers.

This highlights the importance of accurately identifying HER2-low and HER2 ultra-low expression in breast cancer, especially for patients previously classified as HER2-negative. In response, Lunit developed an AI-based whole-slide image (WSI) analyzer for IHC-stained slides to differentiate between true HER2-negative and HER2 ultra-low cases. The AI model evaluated over 67 million tumor cells and 119 million non-tumor cells from 401 WSIs, identifying a significant proportion of HER2 ultra-low cases among pathologist-assessed HER2 score 0 cases.

This AI-powered analysis could expand and refine treatment options for patients with HER2-targeted therapies, as demonstrated by the 23.6% of HER2 score 0 cases identified as HER2 ultra-low by AI, and the 51.9% of HER2 score 1+ cases classified as HER2 low by AI, comparable to the 52.3% objective response rate to a HER2-targeted ADC observed in another clinical trial. In another study, Lunit developed and validated an AI model that analyzes patients' chest CT images alone and in combination with pathology images to predict Immune Checkpoint Inhibitor (ICI) response in NSCLC patients. Lunit's deep learning-based chest CT prediction model, developed using data from 1,876 NSCLC patients treated with ICIs, predicted treatment response based on pre-treatment chest CT scans, along with PD-L1 status and immune phenotype.

The model demonstrated significant predictive power as an independent biomarker. Patients predicted as responders by the AI model showed significantly longer median time to the next treatment (TTNT; 7 months vs. 2.5 months) and a longer overall survival (OS; 16.5 months vs.

7.6 months) compared to patients predicted as non-responders. Combining the AI CT model with histopathologic biomarkers such as PD-L1 expression and tumor-infiltrating lymphocytes (TILs) further enhanced prediction accuracy, highlighting the complementary strengths of imaging and pathology data in improving predictive models for ICI response. A collaborative study with Stanford University School of Medicine examined the association of immune phenotypes with outcomes after immunotherapy in metastatic melanoma, highlighting the heterogeneity of immune phenotypes across melanoma subtypes.

Another study with Northwestern University utilized AI-powered analysis of tertiary lymphoid structures (TLS) in H&E whole-slide images to predict immunotherapy response in NSCLC patients. This demonstrated AI's potential in identifying predictive biomarkers for survival outcomes.