Related Experiment Video
Updated: Oct 11, 2025
07:47
Author Spotlight: Unveiling Transmembrane Protein Family-Related Markers in Gastric Cancer and Implications for Targeted Therapies
Published on: September 15, 2023
1.8K
Integrative multiomics-histopathology analysis for breast cancer classification
Yasha Ektefaie1, William Yuan1, Deborah A Dillon2
1Department of Biomedical Informatics, Harvard Medical School, 10 Shattuck Street, Boston, MA, 02115, USA.
NPJ Breast Cancer
|November 30, 2021
Summary
Deep learning models analyze breast cancer histology images to predict molecular subtypes and mutations. This approach reveals connections between visual features and genetic status, aiding diagnosis and research.
Area of Science:
- Computational pathology
- Biomedical image analysis
- Genomics and transcriptomics
Background:
- Histopathology is crucial for breast cancer diagnosis and subtyping.
- The link between visual morphology and multi-omics data remains underexplored.
Purpose of the Study:
- To develop deep learning models for analyzing histopathology images.
- To investigate the relationship between visual features and molecular characteristics of breast cancer.
Main Methods:
- Weakly supervised deep learning models were trained on H&E-stained breast cancer slides.
- Automated models for tumor detection and subtype classification were developed and validated.
- Model performance was assessed for predicting receptor status, PAM50 status, and TP53 mutation status.
Main Results:
- Automated models achieved high accuracy (AUC ≥ 0.950) in independent cohorts.
- Visual information alone predicted estrogen/progesterone/HER2 receptor status, PAM50 status, and TP53 mutation status.
- Models identified lymphocyte morphology linked to estrogen receptor status and discovered PAM50 genes reflecting cancer morphology.
Conclusions:
- Deep learning image analysis can predict molecular features from histopathology slides.
- This approach bridges the gap between visual morphology and genetic status in breast cancer.
- The models offer utility in both clinical diagnostics and research discovery.
