Advanced Computational Methods in Multi-view Medical Imaging.
详细信息
文摘
Modern automatic imaging methods can help physicians to visualize physical details in the patients body. Such methods must be efficient and reliable to allow making the right decisions during clinical intervention. This thesis presents three medical challenges involving multi-view fluoroscopy and proposes corresponding innovative computational methods. The first method addresses challenges of multiple catheter detection and tracking in fluoroscopic guided interventions. The proposed algorithm incorporates the clinical setup constraints and uses a novel computationally efficient geodesic properties framework that traces the catheter from the sheath to the tip. The method was validated on 1107 fluoroscopic images,demonstrating robust multiple catheter tracking. The second method helps avoid serious complications during a cardiac ablation procedure by extracting esophageal silhouettes and reconstructing a 3D model from fluoroscopic views. The thesis describes the key algorithmic components and suggests how the result could help the clinical workflow. The third method presents automatic image-based cardiac and respiratory cycle gating of image sequences. We developed a novel algorithm for an automatic detection of the moving organ directly from images. The algorithm was validated on X-Ray angiographic and intra-cardiac echocardiography images. The method is very fast and its effectiveness was demonstrated on 3D intra-cardiac ultrasound sequences. All three methods are computationally efficient,robust and represent the state-of-the-art of medical imaging challenges.