Fig. 1

Functions supported by methods and tools described in this paper. Starting from a dataset of whole slide images, a researcher can employ methods in Function 1 to select a subset of images based on image content. If an image has a patch that is similar to the query patch, the image is selected for processing. The selected set of images is then processed through analysis pipelines in Function 2. In the figure, the analysis pipeline segments nuclei in each image and computes a set of shape and texture features. The segmented nuclei and their features are loaded to a database for future analyses in Function 3. In addition, the researcher runs a clustering algorithm to cluster nuclei and patients into groups to look at correlations with groupings based on clinical and genomic data. Clustering, more specifically consensus clustering, requires significant memory and computation power. Using the methods described in this paper, the research can employ a shared-memory system to perform consensus clustering