CIBCB'09 Proceedings of the 6th Annual IEEE conference on Computational Intelligence in Bioinformatics and Computational Biology
| Hi-index | 0.00 |
Histopathology remains one of the most critical steps in diagnosis and treatment of any kind of cancer. Recently, the improvements in imaging techniques led to the discovery of virtual histological slides, which in fact are high resolution images with a resolution of 0.25 μm/pixel. These images enabled researchers to investigate cancer cell morphologies at the finest level. Traditionally, assessment of tumor margins is done manually by a histopathologist. The procedure requires histological processing of the tumor and examining multiple tissue sections. The number of sections varies based on the size, location, and type of tumor. The recognition of cancer areas is time consuming and error prone because of the nature of the human inspecting. Obviously, the use of fast, reliable computerized systems can markedly increase the number of slides that can be examined for the existence of tumor regions. An automated analysis system can benefit the pathologist by automatically examining potentially hundreds of slides per case, which would be a daunting task to perform manually. In this study, in an effort to detect tumors by examining virtual histological slides we introduced a new classification framework. In general, our system learns from subimages that are labeled by a pathologist and attempts to identify areas that contain cancer cells. The main advantage of our framework is to help pathologists make decision on histopathological slides by exploiting high throughput of whole slide imaging systems. When compared with other histopathological image understanding approaches which only search for cell irregularities locally, whole-slide processing exploiting cluster-based features makes our framework unique in this domain.