ESPWA: a deep learning-enabled tool for precision-based use of endocrine therapy in resource-limited settings

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Digital histopathology for cancer diagnostics

Background:

• Cancer morbidity disproportionately impacts low- and middle-income countries (LMICs).
• Delayed or unavailable estrogen receptor (ER) status assessment hinders breast cancer endocrine therapy (ET) in LMICs.
• Empirical ET use leads to ineffective and toxic treatments for ER-negative patients.

Purpose of the Study:

• To develop and validate a deep learning (DL) model, ESPWA (Estrogen Receptor Status Prediction for Haitian patients using deep learning-enabled histopathology Whole Slide Imaging Analysis), for ER status prediction.
• To assess ESPWA's performance directly from hematoxylin-and-eosin (H&E)-stained whole slide images (WSIs).
• To provide an accessible tool for tumor profiling in resource-limited settings.

Main Methods:

• Trained two DL models using weakly supervised attention-based multiple instance learning on H&E WSIs with matched ER status.
• Utilized The Cancer Genome Atlas (TCGA) and Zanmi Lasante (ZL) from Haiti cohorts for model training and validation.
• Evaluated model performance using 10-fold cross-validation and compared against an academic breast pathologist.

Main Results:

• ESPWA demonstrated improved performance (AUROC=0.790) on the Haitian (ZL) cohort compared to a model trained solely on TCGA data (AUROC=0.671).
• ESPWA outperformed an academic breast pathologist in determining ER status from H&E WSIs (accuracy: 0.726 vs 0.639).
• Subgroup analyses identified specific patient populations where ESPWA showed enhanced performance.

Conclusions:

• ESPWA provides an accessible framework for predicting ER status from H&E WSIs, enabling individualized therapeutic insights in LMICs.
• The model's performance surpasses that of expert pathologists in this specific task.
• Clinical trials are underway in Haiti and sub-Saharan Africa to guide precision ET use for prospective breast cancer patients.