excellent accuracy with the therapy response assessed by clinical scores. Furthermore, microscopic aspects allow a more precise assessment of MH. CLE can accurately assess MH in vivo based on the newly developed and statistically validated IBD-MH-score for UC and CC. Mo1640 Computer Assisted Evaluation of Mucosal Microvascular and Crypt Architecture Assessed by Confocal LASER Endomicroscopy in Ulcerative Colitis (UC) Patients With Clinical and Endoscopic Remission Andrea Buda* 1 , Giorgia Hatem 1 , Enrico Grisan 2 , Renata D’Incà 1 , Stefano Realdon 4 , Edoardo Savarino 1 , Claudia Mescoli 3 , ELISA Veronese 2 , Enea Poletti 2 , Giacomo C. Sturniolo 1 1 Surgical, Oncological and Gastroenterological Sciences, University of Padova, Padova, Italy; 2 Information Engineering, University of Padova, Padova, Italy; 3 Pathology, University of Padova, Padova, Italy; 4 Endoscopy, IOV, Padova, Italy Background and Aim: Neoangiogenesis plays a central role in both the initiation and perpetuation of the inflammatory response during chronic intestinal inflammation. However, limited data is available on the microvascular and crypt architecture during remission phases. In this study we have evaluated the intestinal mucosa of UC patients in clinical and endoscopic remission by probe- based confocal endomicroscopy (p-CLE) and quantified microvessel tortuosity and crypt architecture by a semiautomated analysis. Material and methods: 19 UC patients in clinical and endoscopic remission and 19 healthy control subjects were studied. A computer based image processing technique was applied to construct 20 mosaicing images set including 10 colonic crypts from each subject; images were analysed by a dedicated software to evaluate crypt diameter, intercrypt distance, microvessel tortuosity and pericrypt fluorescence. Corresponding biopsies were taken for pathological assessment and UC remitting patients were sub-grouped into either inactive or quiescent disease according to histology. Results: Pericrypt fluorescence (p0.01), crypt diameter (p0.05) but not intercrypt distance (p=0.07) and vessel tortuosity (p=0.13) were significantly increased in UC patients compared to controls. UC patients with inactive disease showed a significant increase in fluorescence leakage (median fluorescence (IQR), 3888 (3560-4240) vs 2696 (2502-3390), p0.01), crypt diameter (median diameter (IQR), 92.5 (85.5-101) vs 73 (70-77) p0.05) intercrypt distance (median distance (IQR), 82.5 (70.5-91.2) vs 66 (59.5-73.5), p0.05) and vessel tortuosity (median (IQR), 1.33 (0.49-1.66) vs 0.27 (0.2-1.01) p0.05) compared to those with quiescent disease. Conclusion: In UC remmiting patients a computer assisted analysis of p-CLE images can detect intramucosal changes of microvessels and crypt architecture in the presence of subtle neangiogenesis and inflammation in endoscopic normal appearing mucosa. Mo1641 Post Food Challenge Confocal LASER Endomicroscopy Reveals Changes Undetected by Histology in Patients With Intractable IBS-Like Symptoms Annette Fritscher-Ravens* 1 , Mark Ellrichmann 1 , Stephan Schoch 1 , Julius Klose 1 , Christoph RöCken 2 , Jochen Brasch 3 , Peter J. Milla 4 1 Experimental Endoscopy, Internal Medicine I, University Hospital Kiel, Kiel, Germany; 2 Pathology, University Hospital Kiel, Kiel, Germany; 3 Dermatology, University Hospital Kiel, Kiel, Germany; 4 Institute of Child Health, UCL, London, United Kingdom Introduction: Confocal laser endomicroscopy(CFLM) is used for real time functional imaging, viewing pathophysiological events in their natural environment in addition to visualizing GI morphology. So far there are no CFLM studies of GI mucosal reactions to food antigens available. Patients with intractable IBS-like symptoms (PII) often perceive that they are induced by food. Studies of the mucosal interaction with food in these patients with CFLM might clarify their response to food antigens. Aim: (i) to detect duodenal mucosal change following challenge with food antigens by CFLM and conventional histology, (ii) to correlate CFLM changes with histological findings in patients with successful dietary treatment following positive CFLM reaction to food. Methods: Consecutive, unselected PII with suspected non-IgE mediated food intolerance were examined by CFLM (Pentax, Hamburg, Germany) in the 2nd part of the duodenum before and after food-challenge. IV fluorescein was used as fluorophore. Five different standardized food antigens (milk, wheat, apple, yeast, soy) were applied sequentially through the endoscope at least 3cm apart from each other. After each provocation the mucosa was observed for 5 minutes. Concentration of intraepithelial lymphocytes(IEL), epithelial breakage with secretion of fluorescein into the lumen and increase in vascularity was assessed. Duodenal biopsies were taken immediately post procedure, H&E and CD3 stained, morphology assessed and IEL’s counted by a blinded independent pathologist. IEL counts 25/100 enterocytes were termed Histo+. CFLM indicated dietary exclusion was followed-up for 6 months Results: 17 patients (mean age 43.6 yrs, 9 male) were studied; in 13 post challenge a marked increase in numbers of lymphocytes was noted on CFLM and fluorescein secretion into the lumen through epithelial breaks and significant increase of vascularity (CFLM+). No change was seen in n=4 (CFLM-). IEL-counts were not significantly different at baseline CFLM-(12,443.79 mean1SD) when compared to CFLM+ (13.152.94), but rose significantly post exposure in CFLM+ (CFLM-: 123.83 vs CFLM+ 17.841.98; p=0.0007). There was also significant difference in IEL counts between Histo- (17.174.8) and Histo+ 37.8016.48; p=0.001). All histo+ were CFLM+, all CFLM- were Histo- but 9/13 CFLM+ were Histo-. Dietary manipulation monitored for 6 months resulted in resolution/marked improvement of symptoms in all CFLM+ patients. Conclusions: These results reflect that food induced mucosal responses proven by dietary follow-up in PII are readily demonstrated by CFLM but poorly shown by conventional histology. These preliminary data need to be confirmed by larger studies. Mo1642 Quantitative Analysis of NBI Magnifying Colonoscopy Images Based on Bag-of-Features for Diagnosis of Colorectal Tumors Yoko Kominami* 1 , Shigeto Yoshida 2 , Shinji Tanaka 2 , Rie Miyaki 1 , Yoji Sanomura 2 , Taiji Matsuo 2 , Hiroyuki Kanao 2 , Shiro Oka 2 , Tsubasa Hirakawa 3 , Tsubasa Mishima 4 , Satoshi Shigemi 4 , Bisser Raytchev 3 , Toru Tamaki 3 , Tetsushi Koide 4 , Kazufumi Kaneda 3 , Kazuaki Chayama 1 1 Gastroenterology and Metabolism, Hiroshima University, Hiroshima city, Japan; 2 Endoscopy, Hiroshima University, Hiroshima city, Japan; 3 Information Engineering, Hiroshima University, Hiroshima city, Japan; 4 Research Institute for Nanodevice and Bio Systems, Hiroshima University, Hiroshima city, Japan Background and Aim: Narrow band imaging (NBI) magnifying colonoscopy can be used to examine various surface mucosal patterns that correlate with the histologic features of colorectal tumors. However, NBI-based diagnosis requires training and expertise. In addition, diagnostic objectivity is needed. Thus, we developed a computerized system to analyze narrow band images quantitatively by means of a local feature-based recognition method, Bag-of-Features, that would yield predictions of histologic diagnoses of colorectal tumors. Objective: To determine whether our newly developed computerized system can predict histologic diagnoses of colorectal tumors depicted on narrow band images. Methods: We obtained NBI magnifying colonoscopy images of 320 neoplastic colorectal lesions that were characterized according to their surface pattern and microvascular architecture (per the previously reported Hiroshima classification system) as type B, n=145; type C1, n=110; type C2, n=28; and type C3, n=37) and diagnosed histologically as adenoma, n=184; carcinoma with intramucosal invasion (M-Ca), n=66; and carcinoma with submucosal invasion (Sm-Ca), n=70. We produced a set of validation images from among these NBI magnifying colonoscopy images. Our custom software uses a Bag-of-Features that recognizes features from regions of interest and classifies regions according to corresponding features. We prepared a set of 966 cut-out images (701 type B images and 265 type C3 images) from another set of NBI images to use as validation images. Local features in the images are represented by dense scale- invariant feature transform (SIFT) descriptors; k clusters (visual words) were generated, and a support vector machine (SVM) classifier with a linear kernel was used to assign numerical attributes (values) to the NBI magnifying colonoscopy images. Regions of interest on NBI magnifying colonoscopy images of the 320 colorectal lesions were classified according to corresponding features on the training images. Results: The SVM output values per Hiroshima classification were as follows: 0.0960.19 for type B, 0.2450.26 for type C1, 0.4900.31 for type C2, and 0.7220.23 for type C3. The differences in values between these classifications were significant. The SVM output values per histologic classification were as follows: 0.1320.05 for adenoma, 0.2660.08 for carcinoma with mucosal invasion (M-Ca), and 0.5640.11 for carcinoma with submucosal invasion (Sm-Ca). The differences in values between these Abstracts www.giejournal.org Volume 77, No. 5S : 2013 GASTROINTESTINAL ENDOSCOPY AB455