Computational Histopathology Pipeline of Tumor Sections

Researchers at LBNL have developed a computational pipeline that downloads newly scanned tumor sections from the NIH TCGA (The Cancer Genome Atlas) repository and then partitions each large formatted image (e.g., 100k-by-100k pixels) into manageable blocks of 1k-by-1k pixels. Using the CITRIS CSE Cluster allows the researchers to generate intermediate files for visualization through panning and zooming, and processing blocks.

The large scale cohort of TCGA has generated a number of hypotheses for tumor subtypes that can be prognostic and/or predictive. These subtypes are being validated on independent datasets. Cluster-computing through CITRIS CSE made large-scale processing for a cohort possible A single tumor section normally takes seven days of computing on a single high-end desktop computer.