AI-powered versus manual HER2 IHC scoring: AI-powered quantification could improve the accuracy of patient selection.
Credit: Nucleai
Breast cancer is the most common cancer for women in the nation, with over 300,000 new cases expected to be diagnosed this year. While the narrative for breast cancer early detection centers around mammograms, the gold standard for diagnosis, clinicians have tapped into enhancing radiology and X-ray imaging using artificial intelligence (AI). This AI-enhanced approach has significantly improved our ability to detect tumors early, changing the lives of many women.
While remarkable, we now understand that the intricacies of breast cancer lie beyond just identifying the tumor with mammograms and understanding the disease at the cellular level. This has led to important discoveries in genetic testing and targeted treatments for specific breast cancer subtypes, such as HER2-positive breast cancer. However, mammograms and genetic tests may not be enough to safely and effectively treat each individual patient.
Because every person’s cancer biology is unique, we need methods that can dig deeper beyond binary detection i.e., biomarker-positive or biomarker-negative. To address these challenges, spatial artificial intelligence (AI), which maps, analyzes, and interprets the complex cellular interactions within breast tumor biopsies can lead to improved clinical decision-making and better-targeted therapeutic interventions beyond conventional diagnostic and screening techniques.
Breast cancer diagnosis and spatial AI
For years, breast cancer has been reduced to a simple, binary “yes or no,” especially for molecular identification of breast cancer subtypes. For example, hormone receptor-positive or negative (HR+/-), estrogen and progesterone receptor positive or negative (ER+/-, PR+/-) and human epidermal growth factor receptor 2 positive or negative (HER2+/-) molecular status have been typically employed to identify a woman’s specific cancer. Based on the presence or absence of these molecular markers, clinicians will determine the most effective therapy for the patient.
This traditional approach to breast cancer diagnosis only detects the presence of a specific tumor marker, positive or negative. However, we now know that this is not enough to determine a patient’s response to therapy. For example, studies have shown that patients who were diagnosed with traditional diagnostics that scored them as a false HER2-negative still responded to HER2-directed antibody-drug conjugate (ADC) trastuzumab deruxtecan treatment. Under current regulatory guidelines, these patients would have been ineligible for HER2-directed ADC treatments. However, in these cases, spatial AI was able to help pathologists determine that patients actually had ultralow expressions of HER2.
Conventional diagnostic approaches often miss such low patterns and complex cellular relationships because the human eye does not always detect cells expressing ultralow levels of HER2 in a background of HER2-negative tumor cells and is also unable to manually compute advanced spatial scores across thousands of cells and hundreds of patient biopsies.
Better precision medicines and clinical trial success with spatial AI
While AI-enhanced mammograms and genetic testing have saved countless lives, many patients still fail to positively respond to treatment or may not receive a specific therapy. This is due to the complexities of cancer biology. Uncovering these complexities through techniques like spatial AI can provide clinicians with a comprehensive understanding of the cellular interactions within the tumor microenvironment, moving beyond traditional diagnostic and genetic approaches. The power of spatial AI is in its ability to correctly categorize patients with biomarker levels too low that could be missed by currently available techniques, and in identifying the subtle patterns that can change treatment outcomes.
Furthermore, spatial AI can provide significant guidance in identifying appropriate patient populations for clinical trial recruitment. More recently, the industry has seen a few clinical cancer trials evaluating ADC-targeted drugs failing to meet their primary endpoints; one of the culprits being that patients were not correctly identified with the appropriate biomarkers. Tapping into AI-driven spatial biomarker diagnostics could help derisk clinical studies and enable biopharma drug developers to improve patient selection for precision medicine trials.
Oscar Puig, Ph.D.
By moving beyond traditional pathology and binary approach to diagnostics, AI-based spatial analysis offers a comprehensive understanding of breast cancer and the full picture of a patient’s tumor biology, ensuring that patients who might otherwise be overlooked receive the most appropriate therapies.
Oscar Puig has worked at Bristol-Myers Squibb, Roche/Genentech, Merck, and Eli Lilly before arriving at Nucleai where he drives strategy in translational medicine and diagnostics. He has over 50 publications in peer-reviewed journals, and his expertise spans all aspects of drug development, from target identification to lifecycle management of approved drugs. He has a special focus on clinical biomarkers and companion diagnostics. Dr. Puig earned his Ph.D. in molecular biology at the University of Valencia, Spain and carried out postdoctoral research at the European Molecular Biology Laboratory in Heidelberg, Germany.
Filed Under: machine learning and AI, Oncology
