Discretized data pattern in endoscopic gastritis images using dynamic window and pairwise gini criterion

Current standard clinical procedure for gastritis is via endoscopy by performing an invasive procedure. The procedure takes tissue samples from patient´s antrum and diagnoses based on pathological evaluation. Several non-invasive computer-aided visualization studies have been conducted to perform feature extraction from the endoscopic gastritis images. Based on an extensive literature search, studies to extract data patterns from the images has never been conducted. Discretization or data pattern extraction is one of the data pre-processing technique that promotes classification. However, data pre-processing is often overlooked by many researchers because it takes up time from the overall classification process. Thus, data pre-processing studies offer faster pre-processing time and compromise with the error rate. Trade-off has been a prolonged issue in discretization studies. Often discretization time is reduced, and the error rate is compromised. However, the proposed discretization algorithm implemented on extracted features from gastritis images has reduced not only the discretization time but also the error rate. As a result of discretization process, it generates good generalization of the data patterns to the endoscopic gastritis extracted features. Thus, determining discretized data patterns from the extracted endoscopic gastritis images may improve the overall classification process in terms of accuracy and learning time.