The long-term goal of this research is to develop a method to quantify the diagnostic image quality of CT images. Image reconstruction plays a pivotal role in any 3D imaging modality, such as CT, MRI, PET, SPECT, etc. As a result, there is ongoing research into developing and optimizing reconstruction algorithms. One limitation of such research is the lack of a quantitative evaluation paradigm that is related/correlated to the diagnostic quality of the image. The goal of this project is to determine the feasibility of using quantitative feature analysis of the reconstructed image as a surrogate for actually measuring radiologists'diagnostic performance on the reconstructed images. Currently methods are qualitative, ad hoc, or quantitative, but not necessarily related to diagnostic quality. When interpreting an image, radiologists use features of lesions in an image to distinguish actual disease from normal anatomy and also to distinguish between different types of pathology. This skill is developed over years of training and experience. We propose to extract quantitative features of lesions to assess the diagnostic quality of a reconstructed image. We will build on over 20 years of experience in extracting and analyzing image features to develop computer-aided diagnosis schemes. We propose to use the feature analysis techniques as a measure of the quality of a reconstructed image. Our hypothesis is that quantitative feature analysis is correlated to diagnostic performance of radiologists. If this is true, then we will have shown that it is feasible to use quantitative feature analysis to evaluate reconstruction algorithms. Specifically in this project, we will develop two databases one containing clinical breast CT images and the other simulated breast CT images with simulated lesions. We will perform an observer study using the clinical images to measure radiologists ability to characterize benign from malignant lesions in images reconstructed using different algorithms. We will use these databases and the observer study to develop a set of quantitative image features that correlate with radiologists'performance in classifying breast lesions. If we are successful, then our method can, with further development, be used to optimize reconstruction algorithms and evaluate dose reduction techniques.

Public Health Relevance

CT reconstruction algorithms are the ultimate factor in image quality and they dictate the final appearance of image that the radiologist will interpret. Unfortunately, no method exists to quantitatively compare reconstruction algorithms in terms of diagnostic image quality (i.e., the ability of radiologists to make the correct diagnosis). This project will determine the feasibility of using quantitative features extracted from reconstructed images to evaluate the diagnostic quality produced from different reconstruction algorithms.

National Institute of Health (NIH)
National Institute of Biomedical Imaging and Bioengineering (NIBIB)
Exploratory/Developmental Grants (R21)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1-BMIT-J (01))
Program Officer
Lopez, Hector
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Pittsburgh
Schools of Medicine
United States
Zip Code
Lee, Juhun; Nishikawa, Robert M; Reiser, Ingrid et al. (2015) Local curvature analysis for classifying breast tumors: Preliminary analysis in dedicated breast CT. Med Phys 42:5479-89