Digital Pathology is a fast-growing field. It enables assessing disease grading, treatment progression, patient prognosis, etc. These analyses are carried out using whole slide tissue images (WSI). Processing of WSIs is costly due to their high resolutions (may reach over 100,000 \(\times\) 100,000 pixels). Modern parallel machines offer adequate computing power to enable fast analyses of WSIs. However, WSI partitioning for parallel load-balanced execution can be hard. This occurs because different areas of a WSI can have distinct/irregular processing costs. We address this problem with a novel algorithm called Background Removal Bisection (BRB), which considers heterogeneity in data processing costs. BRB extends our previous cost-aware data partitioning solution (CADP). The contributions of BRB are: i) a strategy to generate irregular-shaped cost-balanced partitions; ii) effective hierarchical fine-grained background pruning; and, iii) a cost cache that reduces partitioning runtime, essential for large-scale executions. We deployed BRB into the Region Templates runtime and evaluated it with a cancer WSI analysis application. Compared to CADP, BRB achieved speedups of up to \(2.72\times\) in hybrid CPU-GPU machines and up to \(4.5\times\) for 32 compute nodes.
