Detection-theoretic evaluation in digital radiography and optical tomography

Abstract

This dissertation explores the application of objective assessment of image quality (OAIQ) to hardware evaluation for both linear shift-variant and nonlinear imaging systems. We define our task to be the detection of a known signal in either a uniform or structured background. In particular, we study the detection of signals in digital radiography and optical tomography. In digital radiography, current figures of merit are based on stationarity assumptions on the data. The Hotelling observer as we compute it does not make such assumptions. We quantify, from a detection-theoretic perspective, the errors incurred by using stationarity assumptions for nonstationary digital data. We find that by using Monte Carlo methods, the Hotelling observer carries over to the nonlinear setting, and we use it to study spatially varying detectability in optical tomography. In optical tomography there are several data types that can be used to detect signals. Using our methodology, we quantify the information content of those data types. Our results show that information content depends on the type of signal and background as well as how deep the signal is in the tissue. This type of analysis is meant to guide experimental techniques to be suited for the desired detection task.