Computer-Aided Diagnosis of Peritonitis on Cine-MRI Using Deep Optical Flow Network

Abstract

Cine magnetic resonance imaging (MRI) analysis methods are used to diagnose peritonitis, a life-threatening disease associated with decreased intestinal peristalsis. However, variable reproducibility of Cine MRI assessment tests of bowel peristalsis is a critical issue that needs to be improved. Computer-aided diagnosis could help address this issue; however, it faces a number of challenges: the need to extract temporal and spatial features from Cine MRI and the need to extract both global and local features. In this paper, we apply the deep optical flow network (DOFN), an optical flow calculation (TV-L1 method) method based on deep learning for the diagnosis of peritonitis. In the proposed method, the Cine MRI temporal frames serve as input to the DOFN method for optical flow computations of the abdominal region. The computed optical flows are used as global temporal and spatial features for diagnosing peritonitis. In addition, we divide the abdominal region into four subregions. We then proceed with calculating the optical flow and classifying peritonitis in each subregion. The final decision is made based on the results extracted from the four subregions. Furthermore, we visualize the small bowel motility through the computed optical flow. The area under the curve of the proposed method is about 0.72 even without administering contrast media intravenously.