Applying Deep Learning to PD-L1 Expression

Perhaps the greatest promise of Artificial Intelligence (AI) is its potential to enhance our ability to predict, diagnose, and treat disease.

That is why our researchers are continuing to look for new ways to leverage AI to aid our discovery and delivery of effective therapies.

Last week, pathologists from MedImmune and Definiens published a new study relating to AI in the journal Scientific Reports, detailing a novel deep learning solution that enables an automated and objective scoring of PD-L1 expression in late stage non-small cell lung cancer (NSCLC) needle biopsies.

So how exactly can AI improve PD-L1 scoring for NSCLC, and what could it mean for cancer patients?

A Smarter Way to Score PD-L1
PD-L1 expression is a key biomarker for identifying NSCLC patients that may respond to anti PD-1/PD-L1 treatments, making its quantification key to informed treatment decisions.

However, despite the importance of PD-L1 scoring, it’s traditional method of quantification presents a challenge. Current methods of PD-L1 scoring relying on manual microscopic estimations – an imprecise process that allows for scoring variability among pathologists – to determine if patients meet the cut off thresholds for anti PD-1/PD-L1 treatments.

As a result, researchers set out to leverage established deep learning methods, which have the ability to solve complex tasks in areas of image analysis to create an automated, accurate and reproducible method for scoring PD-L1.

By using AI-based methods alongside traditional manual quantification, the team was able to show that the automated algorithm driven methodology was capable of producing precise and reproducible results when compared to traditional methods.

What Improved AI Could Mean
This study demonstrates the greater potential of artificial intelligence when applied to image analysis, and serves as a key proof point for MedImmune’s ongoing study of the tumor microenvironment in collaboration Definiens scientists.

Moving forward, the technical aspects of the image analysis and automated scoring algorithm from this study are being used to explore possible applications as a predictive signature of response for anti PD-1/PD-L1 treatments. With the hope that findings may one day help clinicians make better informed treatment decisions with their patients.

To learn more about our latest study, check out the full paper on Science Reports:

Deep Semi Supervised Generative Learning for Automated Tumor Proportion Scoring on NSCLC Tissue Needle Biopsies