[APPLICATION OF HD-VIDEOENDOSCOPY, MAGNIFYING ENDOSCOPY AND NARROW BAND IMAGING TECHNIQUE FOR THE DIFFERENTIAL DIAGNOSIS OF BENIGN AND NEOPLASTIC EPITHELIAL LESIONS IN THE STOMACH. THE POTENTIAL OF THE COMPUTER-AIDED ANALYSIS].

An endoscopic diagnosis of superficial epithelial neoplastic gastric lesions and early gastric cancer is the challenge of medicine today. It remains at a low level without the use of modern endoscopic technologies such as HDTV and magnifying endoscopy, narrow band imaging (NBI) and similar image-enhanced endoscopic methods, which provide the visualization of microsurface and microvascular pattern. There are a few endoscopic classifications of microsurface and microvascular patterns to distinguish benign and neoplastic superficial gastric epithelial lesions. However, the most effective classifications are based on the intuitive analysis of regularity of surface or/and vascular pattern or heterogeneity of vessels shape and thickness. They are complex for understanding and learning to inexperienced specialists. In this study, we performed expert and computer analysis of 104 HDTV and magnifying NBI endoscopic images of benign and neoplastic gastric lesions in parallel. The images were described for 7 clinical and 23 endoscopic parameters, including 12 qualitative parameters of microsurface and microvascular patterns by the expert evaluation. After statistical analysis, the significant parameters were defined, and the decision rule for the differential diagnosis of benign and malignant lesions were composed. An accuracy of the decision rule was 95.8% for selection of benign lesions and 81.8% for epithelial neoplasia. We performed the computer-aided image analysis using a method "bag of visual words" to distinguish endoscopic images based on irregular vascular pattern as the most significant parameter in expert image analysis and we have shown the accuracy 73-78% for this method. We plan to use this method for independent computer-aided analysis of endoscopic images for differentiation of benign and neoplastic epithelial gastric lesions and creating the clinical decision support system for endoscopy.

[Modern endoscopic diagnosis of precancerous lesions and early cancers of the stomach and colon using computer decision support systems].

Modern endoscopic techniques allow a precise diagnosis of superficial epithelial lesions of the stomach and colon and predict their histological structure. Currently, there are a variety of endoscopic classifications based on the use of magnifying endoscopy and NBI for superficial epithelial lesions according to their morphology. For differential diagnosis of benign lesions, mild neoplasia and early cancer in the colon we commonly use the pit-pattern classification of the surface epithelium created by S. Kudo and mucosal capillary pattern classification created by Y. Sano, which have proven effectiveness in prospective studies. For the stomach to date there is no universally accepted comfortable reliable classification for differentiation benign and neoplastic gastric lesions. However, VS-classification, created by K. Yao, is the most prevalent and effective classification today. It is based on regularity of the vascular and surface (V&S) patterns of the gastric mucosa and presence of the demarcation line on the border with the surrounding mucosa. To increase the efficiency of endoscopic diagnosis with using of these classifications, to identify new diagnostic criteria, to train young specialists and to help skilled doctors computer decision support systems for a physician are successfully developed now.