Abstract
AbstractPurposeInvasive breast cancer patients are increasingly being treated with neoadjuvant chemotherapy, however, only a fraction of the patients respond to it completely. To prevent over-treating patients with a toxic drug, there is an urgent need for biomarkers capable of predicting treatment response before administering the therapy. In this retrospective study, we developed interpretable, deep-learning based biomarkers to predict the pathological complete response (pCR, i.e. the absence of tumor cells in the surgical resection specimens) to neoadjuvant chemotherapy from digital pathology H&E images of pre-treatment breast biopsies.Experimental DesignOur approach consists of two steps: In the first step, using deep learning, mitoses are detected and the tissue segmented into several morphology compartments including tumor, lymphocytes and stroma. In the second step, computational biomarkers are derived from the segmentation and detection output to encode slide-level relationships between the morphological structures with focus on tumor infiltrating lymphocytes (TILs). We developed and evaluated our method on slides from N=721 patients from three European medical centers with triple-negative and Luminal B breast cancers.ResultsThe investigated biomarkers yield statistically significant prediction performance for pCR with areas under the receiver operating characteristic curve between 0.66 and 0.88 depending on the cancer subtype and center.ConclusionThe proposed computational biomarkers predict pathological complete response, but will require more evaluation and finetuning for clinical application. The results further corroborate the potential role of deep learning to automate TILs quantification, and their predictive value in breast cancer neoadjuvant treatment planning.
Publisher
Cold Spring Harbor Laboratory