Department of Computer Engineering
CS 590 SEMINAR
Convolutional Neural Network Cancer Classification in Tissue Images
Troya Cağıl Köylü
Computer Engineering Department
Deep Learning has become a hot topic, mainly due to the limitations of non-complex machine learning algorithms. A type of deep learning primarily manufactured by Yann LeCun et al. and gained recognition with the accomplishments of Alex Krizhevsky and his SuperVision group; Convolutional Neural Networks is considered as the current state of art for image classification. In this study, we apply Convolutional Neural Network learning and classification to distinguish medical tissue images and image regions, regarding their cancer status. Our devised procedures involve 3-class and 6-class classifications of images, alongside with unsupervised evaluation of region division. In addition to Convolutional Neural Network learning and classification, our region division procedure includes image processing tasks of region seed finding and seed region growing, with other assorted tasks. We so far achieved 95% and 84% test set accuracy in 3 and 6-class classifications respectively; and achieved 95%-96%-92% as accuracy, sensitivity, specificity with strictly under 5 region divisions.
DATE: 21 November, 2016, Monday @ 17:00