Automated Tumor Detection in Digital Pathology

TMAs were sectioned and H&E stained on a HE600 platform (Ventana/Roche diagnostics, Oro valley, AZ, USA) using standard diagnostic protocols. Stained TMAs were scanned using an Aperio AT2 system (Leica Biosystems, Wetzlar, Germany) for creation of digital whole slide images (WSIs). WSIs were annotated by a GU-pathologist (HR) using the software QuPath v0.1.2 [21 (link)] as basis of adjustment of automated tumor detection thresholds (JMN). After TMA dearraying and cell detection, smoothed features were calculated for 25 µm and 50 µm. The classifier was trained on tumor and non-tumor regions. The random trees classifier was built with 23,755 training objects and classification results were verified by the pathologist.