Authors

Lin Chen

Type

Text

Type

Dissertation

Advisor

Jiao, Xiangmin | Gindi, Gene | Wu, Song | Zhu, Wei | Liang, Jerome.

Date

2015-12-01

Keywords

CHO Observer, Image Reconstruction, Joint Optimization | Medical imaging and radiology

Department

Department of Applied Mathematics and Statistics.

Language

en_US

Source

This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree.

Identifier

http://hdl.handle.net/11401/77565

Publisher

The Graduate School, Stony Brook University: Stony Brook, NY.

Format

application/pdf

Abstract

In many medical imaging systems used in radiology, the hardware delivers raw data to a sensor and a computer is used to convert the sensor data to a viewable image by implementing tomographic reconstruction algorithms. The radiologist views the image and performs a task, such as detecting a lesion. Engineers and applied mathematicians seek to improve medical imaging systems by making hardware changes and changes in the reconstruction algorithm. Improvement in this thesis is measured by scalar figures of merit, FOM, for task performance. We consider two modalities: SPECT (Single Photon Emission Computed Tomography) and Contrast-Enhanced DBT (Digital Breast Tomosynthesis). For SPECT, the tasks are detection of a signal and detection plus localization of a signal. We seek to improve the collimator (hardware) and regularizer of the reconstruction algorithm to improve SPECT task performance. We use the tools of statistical decision theory. The FOMs are area under the ROC curve for detection and area under the LROC curve for detection plus localization. We find that lower resolution collimators improve performance when coupled with an appropriate regularizer. For Contrast-Enhanced DBT, we alter the sequence of acquisitions by the X-ray tube by interleaving high and low energy acquisitions, and we alter the standard Filtered Backprojection (FBP) algorithm and use OS-SART (Ordered Subset Simultaneous Algebraic Reconstruction Technique). This improves SDNR (Signal-Difference-to-Noise-Ratio), a FOM that is correlated with lesion detection performance. Key to the SPECT work is the use of mathematical observers, essentially feature extraction plus decision algorithms, that effectively emulate human performance in the detection and detection-localization tasks. Without these, the amount of labor involved in human observer studies is unrealistic. | 150 pages

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.