Detection of small bowel tumor based on multi-scale curvelet analysis and fractal technology in capsule endoscopy

Wireless capsule endoscopy (WCE) has been a revolutionary technique to noninvasively inspect gastrointestinal (GI) tract diseases, especially small bowel tumor. However, it is a tedious task for physicians to examine captured images. To develop a computer-aid diagnosis tool for relieving the huge burden of physicians, the intestinal video data from 89 clinical patients with the indications of potential tumors was analyzed. Out of the 89 patients, 15(16.8%) were diagnosed with small bowel tumor. A novel set of textural features that integrate multi-scale curvelet and fractal technology were proposed to distinguish normal images from tumor images. The second order textural descriptors as well as higher order moments between different color channels were computed from images synthesized by the inverse curvelet transform of the selected scales. Then, a classification approach based on support vector machine (SVM) and genetic algorithm (GA) was further employed to select the optimal feature set and classify the real small bowel images. Extensive comparison experiments validate that the proposed automatic diagnosis scheme achieves a promising tumor classification performance of 97.8% sensitivity and 96.7% specificity in the selected images from our clinical data.