| |
| Region of interest segmentation from large histopathology images is an actively researched area given the multitude of applications in pathological research and clinical practice. Here we propose a system to detect regions (objects) of interest in histopathology images using a supervised learning pipeline. Instead of typical k-means in well-used simple linear iterative clustering (SLIC) method, initial superpixel detection is improved by weighted k-means strategy for a better performance at adhering to the object boundary. In each superpixel, multiscale color-texture features are extracted and processed using rolling guidance filters in an effort to reduce inter-class ambiguity and intra-class variation simultaneously. Finally, after feature extraction, a support vector machine (SVM) is trained and applied to segment the testing images. We apply this method to detect pancreatic islets, and in comparison to other approaches, it shows both a dramatic improvement and accuracy compared to existing methods. We envision the system could be used for a variety of other purposes (e.g. tumor detection) in histopathology image analysis. |
|
| Keywords:histopathological image segmentation; supervised learning; rolling guidance filterl; multi-scale features; pancreatic islet |
|
