Detection of epithelial versus mesenchymal regions in 2D images of tumor biopsies using shearlets

摘要

The capability of assessing 2 dimensional images from biological tissue specimens at high resolution requires not only improved optical and biomarker methods, but also is critically dependent on mathematical techniques that enable efficient analyses of larger and more complex data sets in which positional information is accurately assessed. We present here a novel shearlet computational method that detects regions of interest in 2-dimensional tumor biopsy images, using directional information and multiscale analysis. The regions putatively correspond to epithelial or mesenchymal areas of cells, which is of critical interest to clinicians since transition from epithelia to mesenchyme promotes tumor invasion and resistance to chemotherapy. The method significantly outperformed two benchmark methods based on wavelets and shearlets.