Computer-aided delineation of early neoplasia in Barrett's esophagus using high definition endoscopic images

BACKGROUND – Adenocarcinoma of the esophagus is the fastest rising type of cancer in the Western world. The recent development of High Definition endoscopy has enabled the specialist physician to identify Barrett's cancer at an early stage. Nevertheless, it still requires considerable effort, training and expertise to be able to recognize these irregularities associated with early cancer. GOAL – Investigate the technical feasibility of a system that supports the gastroenterologist in finding early Barrett’s cancer. METHODS – An algorithm has been developed for finding early cancer in endoscopic images. We divide the image into small regions, where we quantify color and texture (surface irregularity) information of each region. This information is put into a vector, so we obtain a vector for each region in the image. Next, we employ a Support Vector Machine (SVM) in order to classify each region as being either cancerous, or non-cancerous. The classified regions are used to annotate the early cancerous tissue in the endoscopic image. RESULTS – For training and testing the SVM classifier, 103 images of 30 patients have been selected. We have evaluated different region sizes and color spaces. The SVM achieved a maximum region-based classification accuracy of 94.2% and similar sensitivity and specificity. Figure 1 shows two examples of images annotated by the algorithm compared to their clinical ground truth. CONCLUSION – Our experiments show that our approach is promising for a computer-aided detection system that helps the endoscopist in finding early Barrett’s Cancer. However, further research is needed to investigate the feasibility of such a real-time support system.