Visual Performance in Medical Imaging Using Liquid Crystal Displays.

Abstract

This thesis examined the contrast performance of liquid crystal display (LCD) devices for use in medical imaging. Novel experimental methods were used to measure the ability of medical LCD devices to produce just noticeable contrast. It was demonstrated that medical LCD devices are capable of high performance in medical imaging and are suitable for conducting psychovisual research experiments. Novel methods for measuring and controlling the luminance response of an LCD were presented in Chapter 3 and used to develop a software tools to apply DICOM GSDF calibrations. Several medical LCD systems were calibrated, demonstrating that the methods can be used to reliably measure luminance and manipulate fine contrast. Chapter 4 reports on a novel method to generate low contrast bi-level bar patterns by using the full palette of available gray values. The method was used in a two alternative forced choice (2AFC) psychovisual experiment to measure the contrast threshold of human observers. Using a z-score analysis method, the results were found to be consistent with the Barten model of contrast sensitivity. Chapter 5 examined error distortion associated with using z-scores. A maximum likelihood estimation (MLE) method was presented as an alternative and was used to re-evaluate the results from Chapter 4. The new results were consistent with the Barten model. Simulations were conducted to evaluate the statistical precision of the MLE method in relation to the number and distribution of trials. In Chapter 6, 2AFC tests were conducted examining contrast thresholds for complex sinusoid, white noise, and filtered noise patterns. The sinusoid test results were consistent with the Barten model while the noise patterns required more contrast for visibility. The effects of adaptation were also demonstrated. A noise visibility index (NVI) was introduced to describe noise power weighted by contrast sensitivity. Just noticeable white and filtered noise patterns exhibited similar NVI values. The results indicated that medical LCD devices are capable of high performance for psychovisual testing and presenting complex patterns. However, 256 gray levels are insufficient for high fidelity. 10-bit grayscales can achieve high fidelity, but it is uncertain whether such changes would affect diagnostic performance.