Decision support by fusion in endoscopic diagnosis

In endoscopic image analysis, there are many effective methods to detect the abnormality of an image. However, no individual technique is suitable for detection of any disease pattern in any image. This paper aims to develop a fusion approach to combine multiple techniques to help the physician obtain an accurate diagnosis. Multisensor data fusion technique based on Bayesian Inference is applied in the proposed approach. The combination is based on probability theory and employed nonlinear combination. Before the fusion process, a knowledge-based technique is used for the evaluation of sub-decisions. Similar processed endoscopic case done previously is automatically selected from a case repository and expert physician experience is sought for the supervised evaluation. Meantime, a machine-learning technique is incorporated in the fusion process to increase the accuracy of the decision-making. The new case obtained after the evaluation is fed back as learning data to the fusion process. The proposed decision support approach has been developed. The preliminary results are encouraging and lead support to the feasibility of the method.