On the method-independent lower bound (MILB) to the variance of attenuation estimation in random reflection

Based on the Cramer-Rao inequality, the MILB is calculated for the problem of estimating the frequency slope of the attenuation coefficient of tissue from random reflections of ultrasonic waves. Under typical signal conditions, this bound for a 1-cm x 1-cm region was found to be about 0.08 db/(MHz-cm), rather close to the clinical requirement of 0.1 dB/(MHz-cm). Comparison to existing methods (including an autoregressive deconvolution method) shows room for further improvement.     

Color quantitation through image processing in dermatology

Classical color models and their applications to computer vision are reviewed. The performances of color quantitation from digitized images are compared with those derived from a chromameter. The color quantitation obtained from either digitized color slides or directly digitized images is proved to be more efficient than the conventional visual assessment of observers. A methodology is proposed for determining the specific color indices which are needed in dermatology. An application of this methodology is developed for designing a blanching quantitation index in order to replace the visual assessment during McKenzie tests.     

3-D segmentation of MR images of the head for 3-D display

Algorithms for 3-D segmentation and reconstruction of anatomical surfaces from magnetic resonance imaging (MRI) data are presented. The 3-D extension of the Marr-Hildreth operator is described, and it is shown that its zero crossings are related to anatomical surfaces. For an improved surface definition, morphological filters-dilation and erosion-are applied. From these contours, 3-D reconstructions of skin, bone, brain, and the ventricular system can be generated. Results obtained with different segmentation parameters and surface rendering methods are presented. The fidelity of the generated images comes close to anatomical reality. It is noted that both the convolution and the morphological filtering are computationally expensive, and thus take a long time on a general-purpose computer. Another problem is assigning labels to the constituents of the head; in the current implementation, this is done interactively.