Back to overview

Breast texture synthesis and estimation of the role of the anatomy in the radiological detection process

English title Breast texture synthesis and estimation of the role of the anatomy in the radiological detection process
Applicant Bochud François
Number 104273
Funding scheme Project funding (Div. I-III)
Research institution Institut de Radiophysique Département de Radiologie Université de Lausanne/CHUV
Institution of higher education University of Lausanne - LA
Main discipline Clinical Cancer Research
Start/End 01.08.2004 - 31.07.2006
Approved amount 84'546.00
Show all

Keywords (4)

Image quality; Image texture; Observer model; psychophysical experiments

Lay Summary (English)

Lay summary
Medical image quality is still mainly characterized by subjective methodsbased on visual comparison or the interpretation of images obtained withmore or less realistic test objects. Due to the lack of reproducibilityand confidence of these subjective methods, more objective methods havebeen developed. These were at first used in the case of simple diagnostictasks for images consisting in objects superimposed on white noise orcorrelated noise. The present tendency is to analyze human behaviorregarding anatomical variation and to this purpose, several detectionexperiments in a variety of modalities were performed on real anatomicalbackgrounds. One of the main conclusions of these studies is that theanatomy partly acts as a signal recognizable as such and partly as a purenoise that disturbs the detection process. However, no backgroundparameter (like a statistical moment or a texture parameter) could havebeen isolated as an explanation of this dual role.
In this study, we propose to quantify the role of the anatomy in thedetection process associated with the mammography diagnostic and topredict human performance for realistic detection tasks. These aims couldbe reached in three steps.
First, we want to develop an algorithm that generates realistic anatomicaltextures in 2D projected radiography through the clustered lumpybackground method. This type of background has the crucial advantage ofbeing at the same time realistic and statistically tractable from thealgorithm parameters.
Second, we propose to quantify which statistical and texture parametersare relevant in the detection of a given signal superimposed on ananatomical structure. This could be achieved by analyzing the results ofa series of psychophysical experiments and a detailed and thoroughcharacterization of the image backgrounds by their statistical and textureparameters.
Finally, we plan to develop a model observer that takes the anatomicalcharacteristics into account. This would lead to a better understandingof the detection process performed with real anatomical backgrounds.Moreover, this would allow better optimizing image processing and givingsome hints for the computer assisted diagnostic scientific community.
This work would be mainly conducted in Lausanne at the InstitutUniversitaire de Radiophysique Appliquée. However, it would greatlybenefit from the close collaboration with the Vision & Image UnderstandingLaboratory of University of California Santa Barbara. From the medicalpoint of view, and because this work would be centered on mammography, thecollaboration of Karen Kinkel, MD from Geneva University Hospital and theClinique des Grangettes would guarantee to stay close to the realpractice.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants


Associated projects

Number Title Start Funding scheme
113863 Role of tumor shape and background in breast imaging: from digital mammography to breast tomosynthesis 01.04.2007 Project funding (Div. I-III)