A 3D multi-view freehand ultrasound reconstruction system using volumetric registration and geometric level set segmentation

Freehand 3D ultrasound imaging techniques can be used to reconstruct 3D objects a set of registered 2D image slices. The 2D slices can be located at any arbitrary orientation and position throughout space, and can be acquired using any standard 2D ultrasound transducer in conjunction with an orientation and position sensor. This strategy allows large volumes to be imaged and offers the possibility to upgrade a conventional 2D scanner to a 3D scanner, at a very low cost.; In this dissertation research, I built a multi-view freehand 3D imaging system and demonstrate its effectiveness in reconstructing 3D phantom targets and the left ventricle. The system introduces a number of new features that provide for its improved performance over traditional single-view 3D systems or previously considered multi-view systems. It includes a new ferromagnetic interference detector to guarantee accurate sensor measurements, a new multi-view reconstruction method with automatic volumetric registration, and a new hybrid adaptive gradient vector flow (GVF) geometric active contour (GAC) model for semi-automatic segmentation.; The new system uses a novel method to estimate the probability density functions (PDFs) of position and orientation measurement errors. This method is used to detect electromagnetic interference that can affect the sensor measurements. Using the new detection system, we can guarantee that the system is operating in an interference-free environment, taking accurate position and orientation measurements.; The multi-view reconstruction procedure results in significant reduction in reconstruction error over single view reconstructions. The new volumetric registration approach is performed on binarized walls using non-linear least squares and is shown to be robust to a wide range of initial conditions. We show that the new robust registration method can provide accurate multi-view reconstructions despite significant rigid target motion during different view acquisition.; A new hybrid adaptive gradient vector flow (GVF) geometric active contour (GAC) model is used for single image and image sequence segmentation. The new method can provide significantly improved performance over a competing level set method that was in turn shown to perform better than the original gradient vector flow (GVF) method. It allows for relatively simple and free initialization of the deformable model, while avoiding edge leaking at the poor edges and boundary gaps that are often present in echocardiographic images.; The multi-view system was validated on synthetic data, four ultrasound phantom data sets (each data set with two sequences of 40 or more frames for a total of 336 images) and two echocardiography data sets (each data set has two sequences for a total of 75 images). Volume estimates from multi-view 3D reconstructions were found to be consistently and significantly more accurate than estimates from single-view reconstructions. Furthermore, the results provide us with estimation accuracy as a function of the views and the number of image planes per view. Volume estimates from the 3D multi-view reconstructions with automatic segmentation were found to be in agreement with results from the use of manually segmented images. Using breath-holding and cardiac gating, the system was used to provide single and multi-view 3D reconstructions of the left ventricle at the end of systole and end of diastole phases of the cardiac cycle. Compared to volume estimates from single-view reconstructions, volume estimates from multi-view reconstructions of the left-ventricle were found to be in better agreement with clinical estimates.