Abstract:
HTML
Purpose: A computer-aided diagnosis (CAD) system for the detection of masses on digitized screen-film mammograms (DFMs) was developed in our previous studies. We are developing a mass detection system for mammograms acquired directly by a full field digital mammography (FFDM) system. In this study, we compared the performance of the two systems on pairs of FFDM and DFM obtained from the same patients.
Methods and Materials: Our CAD system consisted of four steps. The input mammogram was first processed with an adaptive density-weight contrast enhancement (DWCE) filter followed by clustering-based region growing to identify suspicious breast structures. Each of these structures was processed by a local refinement stage. Morphological and texture features were then extracted from each detected structure, and rule-based and linear classifiers were trained to differentiate masses from normal tissues. In this study, the mass detection system was adapted to FFDMs by retraining. In an effort to develop a CAD system that is less dependent on the FFDM manufacturer's proprietary preprocessing methods, we used the raw FFDM as input and developed a multi-resolution preprocessing scheme for image enhancement. This scheme consisted of two steps. First, the breast boundary was detected automatically by using Otsu's method. Second, the Laplacian pyramid method was used to decompose the image into multi-scales. A nonlinear weight function based on the pixel gray level from each of the low-pass components was designed to enhance the high-pass components. A data set of 65 cases containing 135 FFDMs acquired with a GE FFDM system and the 135 DFMs of the same view for the same breast was used. The time interval between the DFM and the corresponding FFDM was 0 to 118 days. The data set contained 69 masses. The true locations of the masses were identified by an experienced radiologist.
Results: With initial retraining of the CAD system and preprocessing of the raw FFDM images, our mass detection scheme could perform equally well on the DFMs and the FFDMs. The FROC curves achieved a sensitivity of 80% at 2.1 and 2.1 FP marks/case, respectively, for FFDM and DFM.
Conclusion: Our mass detection CAD scheme can be useful for detecting masses on both FFDMs and DFMs. Further study is underway to improve the various stages of the mass detection system based on the properties of the FFDM images.
Questions about this event email: jvwei@umich.edu
Wei PhD, J,
Computer-aided Diagnosis System for Mass Detection: Comparison of Performance on Full-Field Digital Mammograms and Digitized Film Mammograms. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3105295.html
