Quantifying chromosomal copy number alterations in breast ductal carcinoma in situ: A deep learning based approach

Genomic instability, as measured by chromosomal copy number alterations (CNAs), is associated with progression of ductal carcinoma in situ (DCIS) to invasive breast carcinoma (IBC), and is, therefore, a potential prognostic marker. In this work, we develop a novel image analysis pipeline that utilizes a cascade of biologically salient deep learning models to identify malignant epithelial cells and quantify CNAs. Our automatic approach measures CNAs with a high degree of agreement with those observed by human examiners and performs multiple times faster. It greatly increases the number of cells measured, compared to conventional clinical approaches, and allows for rigorous statistical analysis of specimens. This makes the approach suitable for clinical utility and large-scale studies, not only for breast cancer but also for other types of diseases.