Organ Origin Identification Based on Fine Feature Analysis Using Support Vector Machines
Hongying Lilian Tang · The International Journal of the Computer, the Internet and Management · 2004
With increased usage of digital imaging equipments, approaches that can automatically categorize histological images for patient diagnosis and management are urgently needed more than ever. Since histological images are visually varied and complicated, to automatically identify the organ origin poses difficulties not just for computer but for clinicians as well. In this paper, we propose a hierarchical classification approach: an analysed image is divided into small images that will be measured with predefined meaningful fine features, which are visually discriminated by their texture components. The output of such measurement, although with certain inaccuracy, will be used for organ classification. Multi-class support vector machines (SVMs) are used to implement the hierarchical classification. This two-level classification approach minimises the dependence on accuracy of fine feature detection and considers not only the similarity between fine features but also the relationship between the performance of fine feature classification and organ identification. An empirical research demonstrates good performance of such approach.