Department of Computer Engineering
Bag of subgraphs for Region of Interest Classification in Breast Histopathology
(Supervisor: Assoc. Prof. Dr. Selim Aksoy)
Computer Engineering Department
Breast cancer patients are diagnosed based on biopsy samples taken from patients. Biopsy samples contain regions which play important role when diagnosing. Our method helps pathologists to classify regions of interest (ROI) on biopsy samples. Sizes of these ROIs can vary so it is not straightforward to directly apply Deep Convolutional Networks. Both global and local information in ROIs contain valuable information for classification. By considering that, first we detect potentially informative locations, and represent them as vertices of a graph. Then, an ROI-level graph is constructed by connecting the vertices based on their adjacency relations. Next, an ROI-level representation is formed by counting frequent subgraphs that are learned from graphs in the training data. Finally, the ROI-level representations are used to classify the test samples. Experiments are done by using a data set involving four diagnostic categories.
DATE: 21 October 2020, Wednesday @ 13:35