Delineation of Liver Tumors from CT Scans Using Spectral Clustering with Out-of-Sample Extension and Multi-windowing

Accurate extraction of live tumors from CT data is important for disease management. In this study, an algorithm based on spectral clustering with out-of-sample extension is developed for the semi-automated delineation of liver tumors from 3D CT scans. In this method, spatial information is incorporated into a similarity metric together with low-level image features. A trick of out-of-sample extension is performed to reduce the computational burden in eigen-decomposition for a large matrix. Experimental results show that the developed method using multi-windowing feature obtained better results than using only the original data-depth and the support vector machine method, with a sensitivity of 0.77 and a Jaccard similarity measure of 0.70.