Deep learning-enabled breast cancer hormonal receptor status determination from base-level H&E stains

Area of Science:

• Computational pathology
• Digital pathology
• Artificial intelligence in oncology

Background:

• Estrogen receptor status (ERS) is crucial for breast cancer prognosis and treatment.
• Current ERS determination via immunohistochemistry (IHC) is costly, time-consuming, and prone to variability.
• Hematoxylin and eosin (H&E) staining offers a faster, cheaper, and more consistent alternative for visualizing cellular morphology.

Purpose of the Study:

• To develop and validate a machine learning model for predicting ERS directly from H&E-stained whole slide images (WSI).
• To assess the accuracy of the AI model in determining ERS compared to traditional IHC methods.
• To explore the potential of AI in augmenting cancer diagnosis and treatment planning.

Main Methods:

• Development of a multiple instance learning-based deep neural network.
• Training the algorithm using WSI-level annotations from H&E-stained breast cancer tissue.
• Validation on a diverse, multi-country dataset comprising 3,474 patients.

Main Results:

• The deep neural network achieved an area under the curve (AUC) of 0.92 for both sensitivity and specificity.
• The algorithm accurately determined ERS directly from cellular morphology visible in H&E stains.
• The model demonstrated high performance on a varied, multi-country patient cohort.

Conclusions:

• Machine learning can reliably predict estrogen receptor status from H&E-stained images, offering a potential alternative to IHC.
• This AI-driven approach can enhance the efficiency and accuracy of breast cancer prognosis and theragnosis.
• The technology leverages subtle cellular morphology signals, augmenting clinical decision-making in oncology.