Ultrasound image analysis and recognition techniques for improving workflow in diagnosis and treatment are introduced. Fully automatic techniques for applications of cardiac plane extraction, foetal weight measurement and ultrasound-CT image registration for liver surgery navigation are included. For standard plane extraction in 3D cardiac ultrasound, multiple cardiac landmarks defined in ultrasound cardiac examination guidelines are detected and localized by a Hough-forest-based detector, and by six standard cardiac planes, cardiac diagnosis is extracted following the guideline. For automatic foetal weight measurement, biparietal diameter (BPD), femur length (FL) and abdominal circumference (AC) are estimated by segmenting corresponding organs and regions from foetal ultrasound images. For ultrasound-CT liver image registration, initial alignment is obtained by localizing a corresponding portal vein branch from an intraoperative ultrasound and preoperative CT image pair. Then portal vein regions of the ultrasound-CT image pair are extracted by a machine learning method and are used for image registration.
Part of the book: Medical and Biological Image Analysis