x

    Request a Callback

    Fill in your details and our team will get in touch with you.


    • AACR 2026

    OncoPredikt: A Deep-Learning Framework for Tumor Detection and Biomarker Quantification in Breast Cancer IHC Whole-Slide Images 

    More Author(s)

    San Diego 

    Presented by Dr. Gowhar Shafi et al | April 19, 2026 | Session PO.BCS01.06: Digital Pathology 

    Seeing What Pathologists Miss 

    In breast cancer diagnostics, the stakes of accurate biomarker assessment could not be higher.  

    The expression levels of ER, PR, HER2, and Ki67 measured through immunohistochemistry (IHC) directly determine which treatments a patient receives, and whether they qualify for life-changing therapies such as second-generation antibody-drug conjugates (ADCs). Yet today, this assessment is largely done manually; by pathologists visually examining tissue slides: a process that is inherently subjective, variable, and prone to error. 

    Consider this: the inter-pathologist concordance for distinguishing HER2 0 vs. HER2 1+ stands at just 26%. For patients with HER2-ultralow tumours, a population that now qualifies for ADC therapies:- this diagnostic gap is not just a quality issue. It is a matter of access to life-saving treatment. 

    Introducing OncoPredikt 

    We developed OncoPredikt: an AI-powered deep-learning model built to bring objectivity, consistency, and unprecedented precision to tumour detection and biomarker quantification in breast cancer IHC whole-slide images. 

    Our model was designed and trained on H&E images,130 training samples and 53 validations from TCGA and in-house cohorts:- and tested on IHC whole-slide images for cross-stain generalisation, enabling automated tumour detection and biomarker quantification of ER, PR, HER2, and Ki67 within detected tumour regions. All workflows were validated against pathologist annotations. 

    Results That Redefine the Standard 

    OncoPredikt delivered results that speak for themselves: 

    • Tumour masks achieved Dice Similarity Coefficient >0.8 on H&E images, and even better performance on IHC whole-slide images 
    • For 2 pathologist-analysed IHC 0 (Neg) HER2 samples, our AI approach yielded Ultra-Low (0+) and (1+) predictions:- identifying subtle positivity that visual assessment had missed 
    • The algorithm demonstrated strong concordance with pathologist scoring across ER/PR Allred classification and Ki67 proliferation indices 
    • OncoPredikt showed promising performance in discriminating HER2 0+ ultralow from 0 (no staining):- directly addressing the diagnostic gap where HER2-ultralow patients were previously inaccessible for ADC selection 

    Why This Matters 

    The HER2-ultralow population represents a significant and previously underserved group of breast cancer patients who now qualify for ADC therapies, but only if they are correctly identified.  

    By eliminating observer variability, automating tumour detection, and delivering objective biomarker quantification at scale, OncoPredikt ensures that no eligible patient is misclassified and denied access to the therapies they qualify for. 

    This is AI in precision oncology at its most direct, not just advancing science, but protecting patients. 

    Presented at the AACR Annual Meeting 2026, San Diego, CA | April 17–22, 2026 | #AACR26 
     

    Related Articles

    1Cell.Ai
    Privacy Overview

    This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.