Machine learning models to quantify HER2 for real-time tissue image analysis in prospective clinical trials.

Abstract

3061 Background: Patient eligibility for HER2-targeting treatments is commonly informed by testing tumor HER2 expression using immunohistochemistry. As HER2 expression is visually assessed by pathologists, inter- and intra-rater variability might affect treatment decisions. Here, we report the development of an automated machine learning (ML)-based algorithm to quantify HER2 cell membrane expression across a diversity of breast cancer phenotypes as a clinical tool for monitoring HER2 testing quality. Methods: A total of 689 breast cancer tissue samples were either procured (Avaden Biosciences) or were anonymized samples from the AstraZeneca biobank comprising tissues from primary and metastatic tumors, core needle biopsies and surgical resections, lobular and ductal carcinomas, across tumor grades and HER2 expression levels. Samples were stained for HER2 detection (Ventana HER2 (4B5) Assay) and digitized (Leica Biosystems) across 5 laboratories in the US. Whole-slide images (WSIs) were stratified into training (n = 407), validation (n = 110), and test sets (n = 172). Multiple convolutional neural network based ML models (PathAI, Boston, MA) were trained using 190,000 manual annotations provided by 30 board-certified pathologists to identify artifacts, invasive tumor, identify individual cancer cells and measure tumor cell membrane HER2 expression as partial or complete, and negative, weak-or-moderate, or intense. Cell-level scores were validated against a consensus of manual cell counts from 5 independent pathologists in 320 representative regions of test set WSIs. HER2 scores were generated by automatically applying rules derived from 2018 ASCO/CAP guidelines and then compared in the test set with consensus scores from 3 independent pathologists. Results: Cell counts provided by the ML model were strongly consistent with cell counts obtained by pathologist consensus in all cell-types except for faintly positive HER2 cells where ML-based quantification identified more cells on average. Automatically generated ML-ASCO/CAP HER2 scores using WSI showed substantial consistency across IHC categories with the consensus of pathologists (ICC 0.88, 95%CI 0.82-0.92) in the test set and improved further when ML models were trained to agree with pathologists by adjusting cut offs (ICC 0.91, 95%CI 0.89-0.94). The ML-based model was deployed through the PathAI cloud platform to calculate HER2 testing quality control metrics in real-time in multicentric clinical trials. Conclusions: Automated image analysis of HER2-stained breast cancer tissues using ML-based models is consistent with pathologist consensus across breast cancer tissue types. The results support evidence that ML-based algorithms can help pathologists assess HER2 testing reproducibility in clinical trials.

Cross-institutional HER2 assessment via a computer-aided system using federated learning and stain composition augmentation

“…Despite its aforementioned limitations, the presented HER2 assessment system was able to achieve a comparable performance to existing studies. Glass et al [38] extensively annotated the HER2 expression of individual cancer cells in 689 slides of ductal carcinoma that were stained and digitalized across multiple laboratories, and the resultant model reached an intraclass correlation coefficient (ICC) of 0.91 (95% CI between 0.89 and 0.94) on the 172 testing slides. Wu et al [12] assessed their HER2 scoring algorithm on 246 HER2 0/1+ cases collected by a single medical institution, which achieved an accuracy of 93%.…”

Open questions, current challenges, and future perspectives in targeting human epidermal growth factor receptor 2-low breast cancer

“…More quantitative methods include those that analyze HER2 messenger RNA levels 128 and those that assess HER2 protein using quantitative immunofluorescence or a reverse-phase protein array. 129 , 130 , 131 Additionally, digital pathology and computational methods 122 , 132 , 133 , 134 , 135 , 136 , 137 may help identify HER2-low breast cancer by removing subjectivity and variability and scoring all cells in a sample; they may also define signatures of HER2-low tumors susceptible to HER2-directed therapies. 133 Using samples from the phase I DS8201-A-J101 trial of T-DXd (NCT02564900), a deep learning-based image analysis that quantifies HER2 by generating a quantitative continuous score (QCS) was used to stratify HER2-low metastatic breast cancer into subgroups based on QCS cut-offs that correlated with ORR.…”

An Overview of Clinical Development of Agents for Metastatic or Advanced Breast Cancer Without ERBB2 Amplification (HER2-Low)

“…69,78-80 These techniques may be more accurate than human scoring, and may also offer an advantage by scoring every cell within a sample. 78,79 How-ever, digital imaging may not be available in all institutions; therefore, pragmatic, simple ways to test for ERBB2-low are still needed. More details on these novel assays and computational approaches to ERBB2 scoring are in eAppendix 3 in the Supplement.…”

“…Mass spectrometry has shown promise for assessing ERBB2 status; a multiple reaction monitoring assay yielded more accurate results than IHC and could distinguish between equivocal ERBB2 subgroups that could not be differentiated by IHC . Digital imaging and computational pathology methods (ie, machine learning) could also help distinguish ERBB2 - low from ERBB2 - 0 . These techniques may be more accurate than human scoring, and may also offer an advantage by scoring every cell within a sample .…”

“…Digital imaging and computational pathology methods (ie, machine learning) could also help distinguish ERBB2 - low from ERBB2 - 0 . These techniques may be more accurate than human scoring, and may also offer an advantage by scoring every cell within a sample . However, digital imaging may not be available in all institutions; therefore, pragmatic, simple ways to test for ERBB2 - low are still needed.…”