Study reflects the possibility of a powerful biomarker from pathology data being newly applied to clinical practice to provide actionable insights

Lunit today announced abstract presentation of its findings on the predictive power of AI-based immune phenotype analysis and its correlation with patient survival rate during the American Society of Clinical Oncology (ASCO) Annual Meeting 2020, which will be virtually held beginning on 29 May.

In the abstracts, Lunit presents its findings on the validity of AI classification of tumor microenvironment in non-small cell lung cancer (NSCLC) tissue, deployed to assess the predictive power of AI analysis in ICI (Immune Checkpoint Inhibitors) outcome. This can be critical in identifying additional responders to immunotherapy who otherwise may have been categorized as non-responsive with the current standard-of-care method of identification, using a biomarker called PD-L1.

“Previously pathology tissue slides have mainly been utilized for the diagnosis of cancer, not as a biomarker that can predict treatment response,” said Brandon Suh, CEO of Lunit. “With the advent of technology, these slide images are now being digitalized, creating a new platform for software analysis, of which artificial intelligence can be applied to.”

The study was conducted in partnership with the oncology/pathology department in Samsung Medical Center, Professor Young Kwang Chae from Northwestern University, and Dr. Tony Mok, Professor and Chairman of the Department of Clinical Oncology at the Chinese University of Hong Kong, who is also a member of the Lunit advisory board.

The team employed its AI, Lunit SCOPE, which has been trained with H&E whole-slide images, to analyze the distribution of tumor-infiltrating lymphocytes among NSCLC patients. Tumor-infiltrating lymphocytes are known to be one of the most prominent component of the tumor microenvironment observed to have high predictive value to response to immunotherapy. According to the distribution, the patients were classified by three immune phenotypes—inflamed, excluded, desert. The inflamed type represents an active immune response, excluded represents T-cells being inaccessible to the tumor microenvironment, and the desert type shows no presence of T-cells.

Upon validation of AI-based classification with an independent cohort, patients with the inflamed type, identified by Lunit SCOPE, showed a survival rate seven times higher than that of excluded and desert types. 

Global pharmaceutical companies such as Genentech and Astra-Zeneca have reported through previous studies the correlation between immune phenotypes and the prognosis of immunotherapy in different cancer types. “These studies, however, classifies immune phenotypes according to the observation of the human eye, lacking objective quantification, therefore possessing limitations when it comes to application in clinical practice,” said Suh.

“This study is meaningful as it reflects the possibility of a powerful biomarker from pathology data being newly applied to clinical practice to provide actionable insights,” added Suh. “The analysis of tumor-infiltrating lymphocytes is merely one of the many examples, which can be expanded into diverse analysis when applied to other immune cells. We are planning to broaden our scope of the analysis and accelerate product development.”

In an additional study, the team collected NSCLC patients’ tissue samples from both before and after treatment, classifying it depending on the phenotype mentioned above. The study finds that the observation of the immune phenotype is relevant to the progression-free survival of immune checkpoint inhibitors. For patients with shorter progression-free survival, a change of immune phenotype from inflamed to desert has been observed among half of the patient group.

“This is the first study that used AI in investigating and identifying the immune phenotype in tumor microenvironment,” said Dr. Chan-Young Ock, Vice President of Oncology at Lunit who led the study. “It is a significant step toward developing an AI biomarker for prediction of immunotherapy. Our findings reveal that the immune phenotype of tumor microenvironment would be evolved according to the duration of immune checkpoint inhibitor treatment, suggesting the need for an individually-tailored treatment method, in which AI biomarkers can make a significant contribution.”

It’s not the first time Lunit has presented its studies in ASCO. Last year was the first time, when Lunit’s first proof-of-concept study demonstrated high predictive value of AI-based tissue analysis in predicting response to immune checkpoint inhibitors in lung cancer. This year, the studies have been focused on the immune phenotype.

Based on the study results presented in ASCO this year, Lunit is preparing to expand the AI-based immune phenotype feature into all types of cancers, and development of the AI algorithm as a product. 

“Through this study, we have been able to validate the predictive power of AI-based immune phenotype,” said Kyunghyun Paeng, Chief Product Officer of Oncology Department at Lunit. “With this, we are planning to apply our AI solution to various clinical trials for immunotherapy, finding more evidence as we continue polishing the algorithm to a full-blown software product.”

ASCO 2020 Abstract #3120

Session Title: Developmental Therapeutics—Immunotherapy

Track: Developmental Therapeutics—Immunotherapy

Subtrack: Tissue-Based Biomarkers

ASCO 2020 Abstract #3119 

First Author: Chan-Young Ock, MD

Meeting: 2020 ASCO Virtual Scientific Program

Session Type: Poster Session

Session Title: Developmental Therapeutics—Immunotherapy

Track: Developmental Therapeutics—Immunotherapy

Subtrack: Tissue-Based Biomarkers