Challenges in digital mammography

Chantal Van Ongeval MD
E: [email protected]

André Van Steen
MD

Hilde Bosmans
PhD
Department of Radiology
University Hospitals Leuven Belgium

Full-field digital mammography (FFDM) has definitive  (practical) advantages over screen-film mammography (SFM).  For some systems, the workflow can be markedly increased.  Images are digital; thus, they can be archived  electronically and sent to a remote site for second  ­reading. Computer-aided detection (CAD) can be applied on  the full depth of the data, and dedicated image processing  can improve the perception of the lesions. The performance  of a digital system, however, should ultimately be evaluated  by clinical patient ­studies. The role of the new modality  for screening should be tested in a screening environment.  The first such studies were run on a Senographe 2000D (GE,  Milwaukee, WI, USA). These studies concluded that there was  no difference in detection rate between FFDM and SFM.(1–3)  In the ACRIN (American College of Radiology Imaging Network)  trial, multiple systems of ­different vendors were included.  This trial also concluded that the overall diagnostic  accuracy of digital and film mammography in a screening  environment is similar for the entire population, but that  FFDM is more accurate in women under the age of 50 years.(4)

Technical acceptance tests aim to guarantee an acceptable  level in terms of clinical performance. A first protocol was  recently published by the European Commission.(5) The main  system test discussed in this protocol is the  contrast-detail analysis. This test relies on a test object  with inserts of different size and depth and limiting values  on the minimal thickness (or contrasts) that should be  visible for specified diameters. This test is the subject of  a lot of debate in the literature. There is an important  degree of ­subjectivity ­related to this test, as images  have to be scored manually.
 
The main parameters that characterise the performance of the  detector are: the spatial resolution (or modulation transfer  function: MTF), noise and detective quantum efficiency  (DQE).(6) Currently, the digital detector is still  considered to have a dominant impact on image quality, while  processing protocols and viewing conditions are often not  evaluated.(7) The link between these parameters and clinical  performance remains a challenge today. We are in need of  more information regarding their impact on clinical  performance. One next challenge is concerned with the use of  these modalities in ­screening. How should all digital  advantages be used efficiently?

Clinically and physically, the overall ­conclusion is that  FFDM should become the method of choice in the detection of  breast cancer. But before the method can be considered a  mature ­technology, the challenges, as formulated above,  should be addressed.

The first group of challenges is related to the specific  technical characteristics of FFDM
The requirements on spatial resolution, contrast resolution  and noise of a detector system depend on the radiological  task. Breast cancer can be present as microcalcifications or  masses. Detection of calcifications requires a very  high-resolution imaging technique.(8) Masses induce only  small changes in contrast, and their detection necessitates  high-contrast images. Compared with general radiology,  mammography is a very demanding technique. The spatial  resolution of digital mammography is limited by the pixel  size. FFDM has intrinsically a lower spatial resolution than  SFM. This could cause problems in detecting subtle  microcalcifications. However, a better performance at lower  spatial frequencies seems to compensate for the relatively  large pixel sizes. The apparent contrast resolution is  largely dependent upon the energy of the X-rays and the  image processing that takes place in the ­computer system  after signal acquisition. Differences in X-ray densities can  be more accentuated by post- processing (see Figure 1).  Image processing may react in very different ways on these  anatomical backgrounds. Processing is often driven by the  X-ray distribution in the detector and the texture of the breast. The influence of the X-ray distribution on the result of the applied processing is only rarely investigated. Small studies on the evaluation of image quality showed that processing was not optimal in all patients and for all exposure settings. With inappropriate postprocessing tools, differentiation between noise, calcifications and small densities can be difficult.

[[HHE06_fig1_R16]]

In practice, aspects of spatial resolution, contrast  resolution, noise and processing are interrelated. It is  difficult to predict the outcome of combinations of  characteristics on image quality. Image quality criteria are  necessary for the quantification of the global image  quality. The experience with such criteria remains limited  today and, as far as we know, even criteria for conventional  mammography have not been applied on a large scale, and  limiting values or acceptance levels have not been  proposed.(9)

In digital mammography, there is a shift from hardcopy to  softcopy display and ­reading. Many different display  stations are on the market, but until now there has been no investigation of the effect of a monitor on the reading  results of the ­radiologist. Practical tools are being  developed to control the stability of the monitors.(10)

The second group of challenges is related to screening  mammography
In the screening population, FFDM offers specific  advantages. Integrating FFDM into an existing screening  programme, however, creates a lot of practical problems, in  terms of both informatics and image quality. Different types  of imaging systems are on the market. All of them have other physical charateristics. In addition, most manufacturers have developed image-processing algorithms along with their acquisition system. This results in an always different look to images in the different systems. Images of subsequent runs in a screening programme can therefore appear very variable because of upgrading of the system, changes in the applied processing, etc. Image processing will be vendor-specific, and image quality differences, induced by different quality of processing, will be obvious.
 
Radiologists involved in second reading (ie, reading images  from other mammographic sites) have to deal with all these  different kinds of images (see Figure 2). Can they compare  conventional and digital images of subsequent rounds of the same patient?

[[HHE06_fig2_R17]]
 
For all these reasons, radiologists face an important  learning stage. Dose reduction while maintaining sufficient  image quality is of high importance in a ­screening  population, especially when screening starts at the age of  40 years.(4,11) For a lot of systems, it remains to be  explored at which dose level images should be acquired. The  cost-benefit of a screening programme should not be worse  than with conventional systems. This figure of merit of the  screening programme should be reassessed. As CAD might be  used as a second reader and as this tool has already been  shown to improve cancer detection, application of CAD to all  the images of a screening population is of great interest.  The improvement, however, is dependent on the experience of  the radiologist in mammographic reading and on the type of  lesion (better for calcifications than for masses).(11)

Conclusion
FFDM is already a solution to many of the ­inherent  limitations of screen-film mammography. A series of items  are not fully understood today. The influence of the type of  detectors, processing and display possibilities on the image  and on cancer detection is not sufficiently investigated.  More studies are also needed to explore the optimal dose  levels. Finally, there is only limited experience with  turning an existing screening unit into a fully digital  environment that takes full advantage of the digital nature  of these systems.

References

1. Skaane P, Young K, Skjennald A. Population-based  mammography screening: comparison of screen-film and  full-field digital mammography with soft-copy reading – the  Oslo I study. Radiology 2003;229:877-84.
2. Skaane P, Skjennald A. Screen-film mammography versus  full-field digital mammography with soft-copy reading:randomized trial in a population-based screening program –  The Oslo II Study. Radiology 2004;232:197-204.
3. Lewin FM, Hendrick RE, D’Orsi CJ, Isaacs MF. Comparison  of full-field digital mammography with screen-film  mammography for cancer detection: results of 4.945 paired  examinations. Radiology 2001;218:873-80.
4. Pisano ED, Gatsonis C, Hendrick E. Diagnostic performance  of digital versus film mammography for breast-cancer  screening.N Engl Med 2005;353.
5. European Commission. European Guidelines for Quality  Assurance in Mammography Screening Part B: Digital  mammography. In the Fourth Edition of the European  Guidelines for Breast Cancer Screening. In print by the  European Commission. Downloadable from:www.euref.org
6. Noel A, Thibault F. Digital detectors for mammography:  the technical challenges.Eur Radiol 2004;14:1990-8.
7. Bosman s H, Carton AK, Rogge F, Zanca F, Jacobs J, Van  Ongeval C et al. Image quality measurements and metrics in  full-field digital mammography: an overview. Rad Prot  Dosimetry 2005;117(1-3):120-30.
8. Fischer U, Baum F, Obenauer S, et al.Comparative Study in  patients with microcalcifications: full-field digital  mammography vs screen-film mammography.Eur Radiol  2002;12:2679-83.
9. Van Ongeval C, Bosmans H, Van Steen A. Current challenges  of full field digital mammography.Rad Prot Dosimetry  2005;117(1-3):148-53.
10. Jacobs J, Deprez T, Rogge F, Marchal G, Bosmans H.  Validation of a new dynamic pattern for daily quality  control of medical screen devices. 91st Scientific Assembly  and Annual Meeting of the RSNA, McCormick Place, Chicago,  November 2005.
11. Fischer U, Hermann KP, Baum F. Digital mammography:  current state and future aspects. Eur Radiol 2006;16:38-44.