Search

Analysis of Wireless Capsule Endoscopy (CE) images has become a very active area of research since this novel technology enabled access to previously inaccessible areas of the gastrointestinal tract, especially the small intestine. Art has investigated automatic segmentation of organ boundaries, detection of lesions and bleeding as well as other supervised and unsupervised analysis. However, all of this art has focused on treating the images as individual and independent observations that contribute towards a unique and separate decision. Given the overlap between the images, this is clearly not the case. A human, by contrast, performs assessment by combining the information seen in all neighboring views of the anatomy in a study. This article makes two significant contributions. Towards combining information from multiple images, we propose a supervised classification approach using an HMM framework. Secondly, we use a weak (k-NN) classifier to prototype and evaluate such a framework for regions of the GI tract containing polyps. The combined framework significantly improves the performance of the individual classifier and experiments show promising performance with accuracy > 0.9.

A Wireless Capsule Endoscope (WCE) is a small device that is capable of acquiring thousands of images as it travels through the gastrointestinal track. WCE is becoming a widely accepted method which physicians use in the diagnosis of Crohn's disease, an inflammatory disease that occurs mainly in the small intestine. In this article we present a novel method to detect those images showing inflammation among the thousands of images acquired by the WCE. Further, our method is capable of delineating the inflammation region(s) in each detected frame. Our system utilizes the mean-shift algorithm to find centers of candidate regions that may show Crohn's disease inflammation. Then the system classifies these regions by a trained Support Vector Machine. We have trained, validated and tested our method on three mutually exclusive sets. Our system's testing accuracy, specificity and sensitivity are 87%, 93% and 80% respectively.

Wireless capsule endoscopy (CE) is increasing being used to assess several gastrointestinal(GI) diseases and disorders. Current clinical methods are based on subjective evaluation of images. In this paper, we develop a method for ranking lesions appearing in CE images. This ranking is based on pairwise comparisons among representative images supplied by an expert. With such sparse pairwise rank information for a small number of images, we investigate methods for creating and evaluating global ranking functions. In experiments with CE images, we train statistical classifiers using color and edge feature descriptors extracted frommanually annotated regions of interest. Experiments on a data set using Crohn's disease lesions for lesion severity are presented with the developed ranking functions achieve high accuracy rates.

This paper presents a new approach to evaluation of registration using a general discriminative learning model that is independent of the type of registration method. We select features by association of a registration with a set of metrics (pixel based, patch based and histogram based statistics) and learn a classifier that discriminates mis-registrations from correct registrations using Adaboost. Experiments on a set of wireless capsule endoscopy (CE) images and images extracted from minimally invasive surgical endoscopic video data are presented. Results show that the proposed method outperforms any single classifier.

Capsule endoscopy (CE) is aimed at diagnosing disease in areas of the gastrointestinal (GI) tract beyond the reach of conventional endoscopy. Recent work has addressed various methods for reducing the complexity of CE diagnosis and the time needed for analyzing the data. This includes detection of lumen and its contractions, fluids such as blood and intestinal juices, as well as extraneous matter such as food and bubbles. This paper outlines our ongoing work to segment lesions (in particular Crohn's disease) and other abnormalities in CE images. In particular, here we describe the data collection and clinical analysis for our project and preliminary results for segmenting abnormal and extraneous images from a set of 10 CE studies.