Stochastic Simulations for the Detection of Objects in Three Dimensional Volumes: Applications in Medical Imaging and Ocean Acoustics
Year: 2007
Type: dissertation
Abstract: Given a known signal and perfect knowledge of the environment there exist few\ndetection and estimation problems that cannot be solved. Detection performance is\nlimited by uncertainty in the signal, an imperfect model, uncertainty in environmental\nparameters, or noise. Complex environments such as the ocean acoustic waveguide and\nthe human anatomy are difficult to model exactly as they can differ, change with time,\nor are difficult to measure. We address the uncertainty in the model or parameters by\nincorporating their possibilities in our detection algorithm. Noise in the signal is not so\neasily dismissed and we set out to provide cases in which what is frequently termed a\nnuisance parameter might increase detection performance. If the signal and the noise\ncomponent originate from the same system then it might be reasonable to assume that\nthe noise contains information about the system as well.\nBecause of the negative effects of ionizing radiation it is of interest to maximize\nthe amount of diagnostic information obtained from a single exposure. Scattered\nradiation is typically considered image degrading noise. However it is also dependent\non the structure of the medium and can be estimated using stochastic simulation. We\ndescribe a novel Bayesian approach to signal detection that increases performance by\nincluding some of the characteristics of the scattered signal. This dissertation examines\nmedical imaging problems specific to mammography. In order to model environmental\nuncertainty we have written software to produce realistic voxel phantoms of the breast.\nThe software includes a novel algorithm for producing three dimensional distributions\nof fat and glandular tissue as well as a stochastic ductal branching model.\nThe image produced by a radiographic system cannot be determined analytically\nsince the interactions of particles are a random process. We have developed a particle\ntransport software package to model a complete radiographic system including a\nrealistic x-ray spectrum model, an arbitrary voxel-based medium, and an accurate\nmaterial library. Novel features include an efficient voxel ray tracing algorithm that\nreflects the true statistics of the system as well as the ability to produce separable images\nof scattered and direct radiation.\nSimilarly, the ocean environment includes a high degree of uncertainty. A\npressure wave propagating through a channel produces a measurable collection of\nmultipath arrivals. By modeling changes in the pressure wave front we can estimate the\nexpected pattern that appears at a given location. For this purpose we have created an\nocean acoustic ray tracing code that produces time-domain multipath arrival patterns\nfor arbitrary 3-dimensional environments. This iterative algorithm is based on a\ngeneralized recursive ray acoustics algorithm. To produce a significant gain in\ncomputation speed we model the ocean channel as a linear, time invariant system. It\ndiffers from other ocean propagation codes in that it uses time as the dependent variable\nand can compute sound pressure levels along a ray path effectively measuring the\nspatial impulse response of the ocean medium.\nThis dissertation also investigates Bayesian approaches to source localization in a\n3-D uncertain ocean environment. A time-domain-based optimal a posteriori probability\nbistatic source localization method is presented. This algorithm uses a collection of\nacoustic time arrival patterns that have been propagated through a 3-D acoustic model\nas the observable data. These replica patterns are collected for a possible range of\nunknown environmental parameters. Receiver operating characteristics for a bistatic\ndetection problem are presented using both simulated and measured data.
Source: DukeSpace (Duke University)
Author: Jamie Shorey
Language: English
Subfield: Pulmonary and Respiratory Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W13116202