JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Image analysis-based identification of high risk ER-positive, HER2-negative breast cancers

  • 0Department of Biostatistics, University of North Carolina, Chapel Hill, NC, USA.
Breast Cancer Research : Bcr +

|

December 5, 2024

|

December 5, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

Machine learning on H&E images can identify breast cancer subtypes (Luminal A vs. B) and predict recurrence risk, offering an alternative to genomic testing for ER+/HER2-patients.

Area Of Science

  • Oncology
  • Digital Pathology
  • Machine Learning

Background

  • Luminal A and B breast cancer subtypes are defined by PAM50 gene expression.
  • These subtypes may require different treatment approaches.
  • Machine learning on H&E images may identify features linked to subtype and recurrence risk.

Purpose Of The Study

  • To develop and evaluate an image-based machine learning model for differentiating Luminal A and Luminal B breast cancer subtypes.
  • To compare the performance of the image-based model against genomic testing (PAM50) in predicting recurrence.
  • To assess the utility of image-based methods as an alternative to genomic testing.

Main Methods

  • Pixel-level segmentation of H&E images (n=630 ER+/HER2-cancers) into epithelium and stroma.
  • Feature extraction using convolutional neural networks and multiple instance learning.
  • Training patient-level classification models for Luminal A vs. B discrimination, with and without grade adjustment.

Main Results

  • The image-based protocol differentiated recurrence times with a hazard ratio (HR) of 2.81, comparable to PAM50 (HR=2.66).
  • Grade adjustment improved sensitivity and specificity balance in high-grade tumors.
  • Epithelium-specific and original images showed the highest accuracy.

Conclusions

  • Image-based methods show promise for identifying ER+/HER2-patients who could benefit from further testing.
  • These methods may help overcome low genomic testing uptake rates.
  • Visual classifiers offer a potential alternative to traditional genomic subtyping.

Keywords: