The Robotics and Image Analysis Research Laboratory at UWF

Our research in this area specifically explores the development, utility, and acceptability of computer-based systems than can help to improve radiologists' interpretations of digitized mammogram films. In our system, a digitized version of an X-ray mammogram is processed in a sequence of steps to segment the breast area and enhance abnormal or suspicious regions. The resulting annotated images are then evaluated using ratings by radiologists working at the Ann L. Baroco Center for Women's Health at Sacred Heart Women's Hospital. The images we study are derived from the public database known as the Digital Database for Screening Mammography (DDSM), developed at the University of South Florida. Quantitative assessment of our results is derived from radiologist-marked database ground truth information,

In this research, our goal has not been to make direct comparisons of commercial and academic systems. Alternatively, the focus of this research has been to help to provide a framework for presenting medically relevant issues that can impact the development of digital prompting systems and suggest alternative visualization techniques. Along these lines, we have examined the use of academic outputs alongside outputs provided by the commercially available and FDA-approved R2 ImageChecker M1000 System. As an example, the far left image below shows a digitized mammogram with a malignant abnormality. If detected by the R2 system, a single marker would be placed in this area in the image, prompting a radiologist to examine the area more thoroughly. However, to facilitate such region-of-interest (prompt) understanding, we've found that segmentation images, such as the one shown in the center, appear useful as interpretation aids by enhancing medically relevant features in the prompt areas. Edge-based images such as the one shown on the far right appear useful for facilitating clinical interpretation of possible spicules and for differentiating abnormal and normal tissues. These types of outputs thus have the potential to help mammographers interpret a patient case and guide possible follow-up procedures to fight breast cancer.

Original Image --------- Segmented Image ----- Edge Detected Image

REFERENCES

1. American Cancer Society. Breast Cancer Facts and Figures 1999-2000. Atlanta, GA: American Cancer Society, Inc., 1999.
   
2. Breast Cancer Progress Review Group. Chartering the Course: Priorities for Breast Cancer Research, Executive Summary. Bethesda, MD: National Cancer Institute, 1998.
   
3. Moore SK. Better breast cancer detection. IEEE Spectrum. New York, NY: IEEE Press, 2001:50-54.

• JOURNALS
  • Sentelle, S., Sentelle, C. & Sutton, M.A. (In press). Multiresolution-based segmentation of calcifications for the early detection of breast cancer. Real-Time Imaging: Special Issue on Imaging in Bioinformatics.
  • Sutton, M.A., Sentelle, S. & Rotunda, R. (2002). A pilot study in the analysis of digitized mammograms: Combining academic and commercial results on a public database. Journal of Women's Imaging. 4(1), 7-12.
• CONFERENCES
  • Sentelle, S., Sentelle, C. & Sutton, M.A. (In press). Rapid segmentation of calcification clusters using multi-resolution guided fuzzy c-means and wavelet processing. In IWDM 2002 6th International Workshop on Digital Mammography, Germany: Springer-Verlag Heidelberg.
  • Sutton, M.A., Marin, A., Sentelle S., Wooldridge, S., Griffith, P. & Dell, C. (2001). Development and assessment of protocols for efficient utilization of large-scale digital mammography databases. In Yaffe, M.J. (Ed.), IWDM 2000 5th International Workshop on Digital Mammography, Madison, WI: Medical Physics Publishing, 777-784.