Image Analysis and Enhancement with Applications in Medicine

University dissertation from Dept. of Applied Electronics

Abstract: This thesis consists of six parts in which methods for image analysis and processing are studied. Two-dimensional medical images, image sequences, and three-dimensional image volumes are considered. In the first part, the concept of optical flows (OF) is employed for restoring the contracting/expanding ventricle motion in ultrasonic echocardiograms. Radial velocities obtained by a scalar approach are summed up to time series representing the motion in different sectors of a circular region in the image. In part two, dynamic programming (DP) is used for matching closed deformable contours to the inner boundary of an annular-shaped, bright object in simulated speckle images. Adjustment of method parameters, scale and resolution aspects, search strategies, and spatio-temporal behavior is considered. Application examples consisting of echocardiographic sequences are also given. Part three presents an image enhancement method based on lowpass pyramid decomposition and modified reconstruction. By retaining only the pixels corresponding to detected edges at each reconstruction stage, edge-preserving noise suppression can be obtained. The method is applied to simulated speckle images, to an ultrasonic image from phantom measurements, and to an echocardiographic image. In part four, the algorithm from part three is reformulated to use morphological operations for smoothing before downsampling, reconstruction, and edge detection. Enhancement of speckle image is evaluated. Part five is concerned with the registration of abdominal slices from CT and SPECT using Compton scatter images as a basis for determination of the patient boundary. Contour estimation is performed by interactive thresholding and an automatic deformable boundary technique. In part six, closed boundary finding in medical 3-D image sets is carried out by optimizing Fourier surfaces of gradually increasing order. Algorithmic performance is evaluated for simulated images based on the anthropomorphic Zubal phantom, and various noise levels. Application examples are given for SPECT images acquired from Monte Carlo simulatations and patient measurements.

  This dissertation MIGHT be available in PDF-format. Check this page to see if it is available for download.