SCH: Robust CT Colonography for Local & Cloud-Based Screening

点击翻译按钮获取中文摘要

Colorectal cancer is the third most common cancer in the U.S. It is also the second leading cause of cancer deaths, behind lung cancer. It originates as small growths (polyps) attached to the luminal wall of the colon and rectum. If polyps are not timely diagnosed and treated, they may grow and become cancerous. Untreated colorectal cancer spreads from local invasion of the colon and rectum (in situ) into surrounding tissues, lymph nodes (regional) and eventually to distant parts of the body, e.g., the liver and lungs. If diagnosed early, colorectal cancer has a remarkable recovery rate, reaching over 95%. Therefore, the key for this largely curable disease is early diagnosis and treatment. The American Cancer Society recommends that people at average risk of colorectal cancer start regular screening at age 45. There are four common methods to screen for colorectal cancer: 1) Fecal occult blood test, which detects blood in a stool sample that is not visible; 2) A fecal immunochemical test, which detects occult blood in stool; 3) Optical Colonoscopy, where a flexible endoscope is inserted to visually inspect the interior walls of the rectum and colon; and 4) Computed Tomography Colonography, which remotely visualizes the interior of the colon using a 3D reconstructed model of the colon from an abdominal CT scan of prepped patients. Coordination of Computed Tomography Colonography (for polyps detection and classification) and Optical Colonoscopy (for validation and removal of polyps) holds the best option to detect and prevent cancer. This NSF-SCH project deals with using Computed Tomography Colonography as a non-invasive early screening and follow-up for colorectal cancer, and would research and create methods for optimizing it and synchronization with Optical Colonoscopy. From a computational perspective, Computed Tomography Colonography involves five steps to analyze a patient-prepped abdominal CT scan: 1) image processing (e.g., Electronic Colon Cleansing) to correct prep and scanner errors; 2) image segmentation to isolate the colon tissue from the rest of the abdomen; 3) 3D reconstruction to generate a volumetric representation of the colon; 4) visualization of the luminal surface generated by the 3D model for polyp detection and assessment; and 5) analysis to catalog detected polyps location, size, shape, and potential pathology. This NSF SCH project has three goals: 1) Establish an analytic approach for the entire pipeline of the Computed Tomography Colonography system, to augment the published and patented progress made by the investigators in the visualization step; 2) Develop an optimal implementation of the newly discovered Fly-In visualization approach, which uses a rig of virtual cameras to navigate inside the 3D model to enable expert and automatic polyp detection; and 3) Develop a front-end Computed Tomography Colonography system that lends itself to human and artificial intelligence-based reading of massively large number of Computed Tomography Colonography scans locally and on the cloud, from widely distributed geographical locations. Novelties to be explored and implemented include: 1) use a combination of Markov Random Field and Deep Learning for automatic segmentation of the colon from the abdominal CT scan of prepped patients; 2) Registration of supine and prone CT scans using deformable models and discrete optimization; 3) Optimization of the newly discovered Fly-In visualization method, in terms of the number of virtual cameras used for visualization, proper representation of the lumen surface, alternating projections of 2D CT scans (axial, sagittal and coronal) and images in the field of view of the camera rig, and discrimination of polyps with respect to type, size and location in the lumen; 4) Create optimal detection and classification algorithms for small-size precancerous polyps in the Fly-In approach using novel machine learning techniques; 5) Design a robust cloud-based Computed Tomography Colonography reading system which allows for local reading by expert radiologists and on the cloud, using sufficient cases by renowned experts, in order to provide a measured impact of Computed Tomography Colonography for diagnosis of colorectal cancer. These tasks include novel theoretical and computational methods, collaboration of a large multidisciplinary team, and will provide a fertile environment for training of graduate students and biomedical researchers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.