NASA Cooperative Agreement # NCC 2 - 881 " Advanced Signal Processing Methods Applied to Digital

The multiresolution wavelet expansion of digitized mammograms can be analyzed using a parametric statistical model for each image of the expansion. The statistical analysis of the individual expansion components is relatively simple, whereas the analysis of the original image is complicated. An important application of this technique is the statistical modeling of normal tissue in digital mammograms. One possible application of this analysis is to the identification and separation of normal tissue from calcified tissue. The multiresolution probability modeling can be generalized and applied to other digitized medical images, or to any digital image where rigorous statistical evaluation is appropriate. This work was supported in part by the National Aeronautics and Space Administration under Grant NASA NCC 2-881, Ames Research Center, Mountain View, California. John J. Heine is with the Digital Medical Imaging Program, Department of Radiology, University of South Florida, Tampa, FL 33612-4799 (e-mail: [email protected]). Stanley R. Deans is with the Department of Physics, and Member H. Lee Moffitt Research Center, University of South Florida, Tampa, FL 33620-5700, (e-mail: [email protected]). D. Kent Cullers and Richard Stauduhar are with the SETI Institute, 2035 Landings Drive, Mountain View, CA 94043, (e-mail: [email protected], [email protected]). Laurence P. Clarke is with the Department of Radiology, and It. Lee Moffitt Research Center, University of South Florida, Tampa, FL 33612-4799 (e-mail: [email protected]). Multiresolution Statistical Analysis of High Resolution Digital Mammograms Part II: Application John J. Heine. Stanley R. Deans, Senior Member, IEEE, D. Kent Cullers, Richard Stauduhar. Member IEEE, and Laurence P. Clarke, Member IEEE AbstractA multiresolution statistical method for identifying clinically normal tissue in digitized mammograms is used to construct an algorithm for separating normal regions from potentially abnormal regions; that is, small regions that may contain isolated calcifications. This is the initial phase of the development of a general method for the automatic recognition of normal mammograms. The first step is to decompose the image with a wavelet expansion that yields a sum of independent images, each containing different levels of image detail. When calcifications are present, there is strong empirical evidence that only some of the image components are necessary for the purpose of detecting a deviation from normal. The underlying statistic for each of the selected expansion components can be modeled with a simple parametric probability distribution function. This function serves as an instrument for the development of a statistical test that allows for the recognition of normal tissue regions. The distribution function depends on only one parameter, and this parameter itself has an underlying statistical distribution. The values of this parameter define a summary statistic that can be used to set detection error rates. Once the summary statistic is determined, spatial filters that are matched to resolution are applied independently to each selected expansion image. Regions of the image that correlate with the normal statistical model are discarded and regions in disagreement (suspicious areas) are flagged. These results are combined to produce a detection output image consisting only of suspicious areas. This type of detection output is amenable to further processing that may ultimately lead to a fully automated algorithm for the identification of normal mammograms. A ground truth evaluation of the merit of this method reveals that reasonable predictions of isolated false positives is possible prior to detection, and a specificity of 46% can be maintained while keeping the sensitivity at 100%.A multiresolution statistical method for identifying clinically normal tissue in digitized mammograms is used to construct an algorithm for separating normal regions from potentially abnormal regions; that is, small regions that may contain isolated calcifications. This is the initial phase of the development of a general method for the automatic recognition of normal mammograms. The first step is to decompose the image with a wavelet expansion that yields a sum of independent images, each containing different levels of image detail. When calcifications are present, there is strong empirical evidence that only some of the image components are necessary for the purpose of detecting a deviation from normal. The underlying statistic for each of the selected expansion components can be modeled with a simple parametric probability distribution function. This function serves as an instrument for the development of a statistical test that allows for the recognition of normal tissue regions. The distribution function depends on only one parameter, and this parameter itself has an underlying statistical distribution. The values of this parameter define a summary statistic that can be used to set detection error rates. Once the summary statistic is determined, spatial filters that are matched to resolution are applied independently to each selected expansion image. Regions of the image that correlate with the normal statistical model are discarded and regions in disagreement (suspicious areas) are flagged. These results are combined to produce a detection output image consisting only of suspicious areas. This type of detection output is amenable to further processing that may ultimately lead to a fully automated algorithm for the identification of normal mammograms. A ground truth evaluation of the merit of this method reveals that reasonable predictions of isolated false positives is possible prior to detection, and a specificity of 46% can be maintained while keeping the sensitivity at 100%. This work was supported in part by the National Aeronautics and Space Administration under Grant NASA NCC 2-881, Ames Research Center, Mountain View, California. John J. Heine is with the Digital Medical Imaging Program, Department of Radiology, University of South Florida, Tampa, FL 33612-4799 (e-mail: [email protected]). Stanley R. Deans is with the Department of Physics, and Member It. Lee Moffitt Research Center, University of South Florida, Tampa, FL 33620-5700, (e-mail: [email protected]). D. Kent Cullers and Richard Stauduhar are with the SETI Institute, 2035 Landings Drive, Mountain View, CA 94043, (e-mail: [email protected], [email protected]). Laurence P. Clarke is with the Department of Radiology, and H. Lee Moffitt Research Center, University of South Florida, Tampa, FL 33612-4799 (e-mail: [email protected]). SETI Institute and The University of South Florida at Moffitt Cancer Research Center Digital Mammography: Multiresolution Statistical Methods for Normal Image Recognition Richard P. Stauduhar SETI Institute D. Kent Cullers SETI Institute Stanley R. Deans Department of Physics USF Laurence P. Clarke Department of Radiology USF John J. Heine Ph.D. Student USF Copy of slides used by D. Kent Cullers in his presentation on May 8, 1996, at the conference: Aerospace Medical Association 67th Annual Scientific Meeting Medical Applications of Space Research and Technology Atlanta, Georgia, May 6-9, 1996.