Gene expression-based recurrence assays are strongly recommended to guide the use of chemotherapy in hormone receptor-positive, HER2-negative breast cancer, but such testing is expensive, can contribute to delays in care, and may not be available in low-resource settings. Here, we describe the training and independent validation of a deep learning model that predicts recurrence assay result and risk of recurrence using both digital histology and clinical risk factors. We demonstrate that this approach outperforms an established clinical nomogram (area under the receiver operating characteristic curve of 0.83 versus 0.76 in an external validation cohort, p = 0.0005) and can identify a subset of patients with excellent prognoses who may not need further genomic testing.
Xception-based deep learning models were trained on 1,039 patients from TCGA to allow for unsupervised predictions on external data. One model was trained to identify image tiles within pathologist annotation of tumor versus background image tiles. The second model was trained to predict a research version of the 21-gene recurrence score calculated from gene expression data from the annotated tumor regions from TCGA. Finally, a combined clinical / pathologic model was developed by fitting a logistic regression to deep learning model predictions and the University of Tennessee clinical nomogram predictions.