POS1391 DEEP LEARNING-BASED PANNUS LOCALIZATION SOFTWARE IN THE HANDS OF INFLAMMATORY ARTHRITIS USING TIME-INTENSITY CURVE (TIC) SHAPE CLASSIFICATION ON DYNAMIC MRI DATASET

Background Rheumatoid arthritis (RA) is a chronic inflammatory disease associated with significant functional impairment and disability, linked to structural articular peri-articular damage [1]. Synovitis characteristic feature of RA, considered an important factor in activity the best predictive marker joint [2]. Therefore, accurate quantification synovitis can play role clinical evaluation treatment serving as biomarker. Pixel-by-pixel time–intensity curve (TIC) shape analysis dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) technique help visualize differently shaped TICs [3]. Pixels having TIC type 4, which characterized early enhancement followed by washout phase, were regarded pixel [4]. A deep residual network called ResNet-50 model, convolutional neural (CNN) that 50 layers deep, might effectively classify shapes. Objectives This study aimed develop software automatically demonstrate distribution enhancing synovial pannus patients on DCE-MRI. Methods Modified was used MATLAB. Two investigators drew regions interest (ROIs) muscle, bone, under expert guidance obtained for each tissue DCE-MRI ten rheumatoid patients. Through cross-validation batch evaluation, we verified confusion matrices performance classifier. Then identified pixels image classification from pixel-by-pixel displayed them color. The evaluated using visual assessment seven hand joints one patient. Results 150 ROIs muscles, bones, 59 drawn 4049, 3825, 1041 images, respectively. classifier’s accuracy, precision, sensitivity, specificity 99.6%, 99.3%, 98.4%, 99.7%, Out joints, four assessed good, two fair, fail. Figure 1. Software contrast enhanced MRI wrist Conclusion Our classifier showed high specificity. And through comparison between manual outlining result software, our had relatively good performance. automatic developed learning CNN accurately display RA. References [1]Micu MC, Fodor D. Concepts monitoring arthritis- musculoskeletal ultrasound. Part I: synovitis. Med Ultrason. 2015;17(3):367-76. [2]Carotti M, Galeazzi V, Catucci F, Zappia Arrigoni Barile A, et al. Clinical utility eco-color-power Doppler ultrasonography interpretation synovitis: review. Acta Biomed. 2018;89(1-s):48-77. [3]Sakashita T, Kamishima Kobayashi Y, Sugimori H, Tang Sutherland K, Accurate quantitative pixel-by-pixel, time-intensity analysis. Br J Radiol. 2016;89(1061):20151000. [4]Kobayashi Ichikawa S, Noguchi Kono Quantification arthritis: Arterial mask subtraction reinforced mutual information improve accuracy MRI. Magn Reson Imaging. 2018. Disclosure Interests None declared