Model-based image processing techniques for breast cancer detection in digital mammography

The objective of the research is the development of hybrid schemes using model-based image processing techniques for the detection of microcalcifications and masses in digital mammography. Microcalcifications and masses are important signs leading to early breast cancer detection.;In this research, we develop a methodology based on fractal modeling to analyze and simulate breast background structures. Therefore, microcalcifications can be enhanced by taking the difference between the original image and the modeled image. The results demonstrate that the fractal modeling method is an effective way to enhance microcalcifications, and in turn can facilitate the radiologist's detection of microcalcifications. We also apply a convolution neural network (CNN) classifier to the automated detection of clustered microcalcifications. We propose a partial wavelet reconstruction approach to enhance the signal pattern in the regions of interest (ROIs) which are located after the prescan. The system performance is significantly improved after using the enhanced ROIs as the input to the CNN.;For mass detection, we develop a combined method utilizing morphological operations, a finite generalized Gaussian mixture (FGGM) modeling, and a contextual Bayesian relaxation labeling technique (CBRL) to enhance and extract suspicious masses. We use a multi-modular neural network (MMNN) to distinguish true masses from non-masses based on the features extracted from the suspected regions. The results demonstrate that all the areas of suspicious masses, as well as high contrast tissues, in mammograms are extracted in the prescan step using the proposed segmentation procedure. The MMNN can reduce the number of suspicious regions and identify the true masses. All experimental results indicate that morphological filtering combined with the FGGM model-based segmentation is an effective way to extract mammographic suspicious mass patterns. Compared with conventional neural networks, the MMNN can lead to more efficient learning algorithm and can provide more understanding in the analysis of the distribution patterns of multiple features extracted from the suspicious masses.