Information theoretic design of breast sonography

Breast sonography;ideal observer;image quality;spatial filters

Nguyen, Nghia Q. ; Insana, Michael F. ; Insana ; Michael F.

关键词: Breast sonography;    ideal observer;    image quality;    spatial filters;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/14624/Nguyen_Nghia.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

First-principle approaches to the design of medical ultrasonic imaging systems for specific visual tasks are being explored. Our study focuses on breast cancer diagnosis and is based on the ideal observer concept for visual discrimination tasks, whereby tasks based on five clinical features are expressed mathematically as likelihood functions. Realistic approximations to the ideal strategy for each task are proposed as an additional beamforming procedure to maximize diagnostic image information content available to readers.Our previous study revealed that the Wiener filter, derived as a stationary approximation of the ideal observer and operating on RF echo data, generally improved discriminability except for one case involving high-contrast lesions.This study explores an adaptive, iterative Wiener filter that includes a lesion segmentation algorithm to improve discriminability for high-contrast lesions. Predicted performance is compared with that measured from trained human observers using psychophysical methods. The iterative Wiener filter was found to match the performance of the Wiener filter for low-contrast lesions and increase the performance for high-contrast tasks. This filter offers greater diagnostic performance for discriminating malignant and benign breast lesions, and it provides a rational basis for further task-specific imaging system design. The thesis also addresses some issues that may be encountered when applying the filter in clinical environments. We found that background variability improves the performance of spatial filters. We also studied the limitation of these filtering methods as key assumptions are violated.