Computerized analysis of images in the detection and diagnosis of breast cancer
Section snippets
Image quality and CAD
Signal detection theory applies to all observers, whether they are human or computer. Therefore, one would not expect a CAD system to be able to detect a lesion having low signal levels or in high noise levels on poorly obtained mammograms.12 Conversely, one would anticipate that the performance of a computerized analysis of mammograms would improve if the acquired breast images were of high image quality. Improved signal to noise characteristics, as expected with full-field digital mammography
Computer vision and artificial intelligence
Computer vision involves the automatic extraction of features, which may or may not be visible to a human observer, from a digital image. The development of computer vision methods requires a priori knowledge of the physical imaging properties of the digital image acquisition system and the morphological information concerning the abnormality (e.g., mass lesion or cluster of microcalcifications) along with the normal parenchyma. A sufficient number and types of cases in a database is needed to
Computer-aided detection
Computerized detection of lesions involves having the computer locate suspicious regions, leaving the subsequent classification (e.g., probability of malignancy status) of the lesion and patient management decisions to the radiologist. In such situations, the computer is acting as a second reader, like a spell checker, in the breast cancer screening process. Most computerized detection schemes in mammography are being developed for the detection of mass lesions or clustered microcalcifications.
CAD in diagnostic breast imaging
Once a possible abnormality is detected, it characteristics must be evaluated by the radiologist in order to estimate a likelihood of malignancy and to yield a decision on patient management. Characteristics of the lesion may be evaluated further by multiple imaging techniques, including special view mammography, ultrasound, and magnetic resonance imaging (MRI), in order to improve the positive predictive value for biopsy recommendations.
The initial investigations into the use of computers in
Indices of similarity and human/computer interface
Radiologists interpret cases rather than individual images, thus the computerized analysis of breast images can be case-based as opposed to image-based. Computer analysis of multiple views or multiple modalities, however, requires effective and efficient displays in order to communicate the multiple images and output to the radiologist. Research has been performed to develop display interfaces, which would better present the computer output to the radiologist.74, 75, 76, 77
The output of a
Summary
Limitations in the human eye-brain visual system, the presence of overlapping structures in images, and the vast umber of normal cases in screening programs provide motivation for the use of computer techniques that have the potential to improve detection and diagnostic performance, and ultimately patient care. The success of computerized analyses of breast images depends on both the ability of the computer vision and artificial intelligence techniques that extract and characterize suspect
Acknowledgements
The author is grateful for the many fruitful discussions she has had with the faculty and research staff in the Department of Radiology, University of Chicago, especially to Carl J. Vyborny, MD, PhD, who passed away during the preparation of this manuscript. M.L. Giger is a shareholder in R2 Technology, Inc., Los Altos, CA. It is the University of Chicago conflict-of-interest policy that investigators disclose publicly actual or potential significant financial interests that may appear to be
References (77)
- et al.
Computer-aided detection and diagnosis
Radiol Clin North Am
(2000) - et al.
Prescreening entire mammograms for masses with artificial neural networksPreliminary results
Acad Radiol
(1997) - et al.
Computer-assisted detection of interval breast cancers
Eur J Radiol
(2001) - et al.
Improving breast cancer diagnosis with computer-aided diagnosis
Acad Radiol
(1999) - et al.
Computerized analysis of lesions in us images of the breast
Acad Radiol
(1999) - et al.
Improving the distinction between benign and malignant breast lesionsthe value of sonographic texture analysis
Ultrason Imaging
(1993) - et al.
Performance of CAD in the interpretation of lesions on breast sonography
Acad Radiol
(2004) - et al.
Discrimination of MR images of breast masses using fractal interpolation function models
Acad Radiol
(1999) - et al.
Variability in the interpretation of screening mammograms by U.S. radiologists
Arch Intern Med
(1996) - et al.
Variability in radiologists’ interpretations of mammograms
New Engl J Med
(1994)