MICCAI 2019 Wednesday

MICCAI 2019 DAILY 8 Oral Presentation Ruibin Ma is a fourth-year PhD student from the University of North Carolina at Chapel Hill, working onmedical image analysis and computer vision under the supervision of Professor Stephen Pizer. He talks to us ahead of his oral and poster session this afternoon. Ruibin tells us that during colonoscopy , the physician can miss some parts of the surface of the colon and this could cause polyps to be missed. Polyps are lesions in the colon that have to be eliminated when the physician sees them as they may be pre-cancerous. From a first-person perspective, it's not always possible to tell which parts are missing. Especially when the surface is obscured by a haustra ridge, which are small rings in the large intestine that can block the surface behind. If areas are missed, that could potentially be a major problem. They detect those missing surfaces by reconstructing a 3D surface in real-time from endoscopy videos. Away from the medical field, self- driving cars use a robotic technology called SLAM to create a 3D world from monocular video signals. They do the same in colonoscopy. In this 3D colon, whatever is not seen, they leave as blank holes. If a large blank region is seen, an alert is set for the physician to tell them that they may want to revisit that position because they have missed something. It supports them to have a larger view of what is happening in the real organ of the patient and provides useful feedback and suggestions online during operations. Ruibin says their technology is the first to reconstruct this dense 3D surface in real time. Most techniques don't work well as colon images are so textureless. By contrast, in the real world, there are corners of buildings and many salient points for common computer vision methods to detect, but the colon doesn't have that. They resolve this by combining a deep neural network with a Real-time 3D reconstruction of colonoscopic surfaces for determining missing regions