Automated Quality Assurance Applied to Mammographic Imaging
- PDF / 983,400 Bytes
- 10 Pages / 600 x 792 pts Page_size
- 82 Downloads / 202 Views
Automated Quality Assurance Applied to Mammographic Imaging Lilian Blot School of Information Systems, University of East Anglia, Norwich NR4-7TJ, UK Email: [email protected]
Anne Davis Portsmouth Hospitals NHS Trust PO3 6AD, Portsmouth, UK Email: [email protected]
Mike Holubinka Portsmouth Hospitals NHS Trust PO3 6AD, Portsmouth, UK Email: [email protected]
Robert Mart´ı School of Information Systems, University of East Anglia, Norwich NR4-7TJ, UK Email: [email protected]
Reyer Zwiggelaar School of Information Systems, University of East Anglia, Norwich NR4-7TJ, UK Email: [email protected] Received 31 July 2001 and in revised form 13 March 2002 Quality control in mammography is based upon subjective interpretation of the image quality of a test phantom. In order to suppress subjectivity due to the human observer, automated computer analysis of the Leeds TOR(MAM) test phantom is investigated. Texture analysis via grey-level co-occurrence matrices is used to detect structures in the test object. Scoring of the substructures in the phantom is based on grey-level differences between regions and information from grey-level co-occurrence matrices. The results from scoring groups of particles within the phantom are presented. Keywords and phrases: automatic quality control, mammographic images, grey-level co-occurrence matrices, image segmentation.
1. INTRODUCTION Meticulous quality control of mammography was advocated [1] as a mandatory requirement of the National Health Service Breast Screening Programme prior to the establishment of the NHSBSP. The quality control is with respect to evaluating system performance and for long term monitoring. Standards of acceptability have been established based upon subjective interpretation of image quality test phantoms, for example, Leeds test objects TOR(MAM), TOR(MAX), or TOR(MAS). Leeds TOR(MAM) is widely used for this purpose, and has been reported as being among the most consistent and sensitive [2] of those available commercially. This study is aimed at applying image analysis methods that will be employed for the analysis of phantom films
(which were digitised) for quantitative objective assessment of different imaging systems, monitoring temporal performance, and evaluating the impact of introducing new imaging systems, for example, direct digital mammography. Analysis of phantom images relies upon scoring by experienced observers according to the visibility of details. Scoring schemes have been proposed whereby the observer allocates a value according to visibility of the test details (i.e., 0 = detail not seen, 1 = barely visible/threshold, 2 = less visible/faint, 3 = detail easily seen). Experienced observers have shown good self-consistency but have yielded substantially different scores when scoring images from the same phantom and when using the same protocol [3]. These interobserver variations clearly have the potential to call into question the
Automated Quality Assurance Applied to Mammographic Imaging
737
Image Calcium-based p
Data Loading...
