Self organizing map neural network with fuzzy screening for micro-calcifications detection on mammograms
Chui‐Mei Tiu, Tai‐Lang Jong, Chi‐Wen Hsieh · 2008
Mammography remains the main screening tool for detecting breast cancer. Depicting micro-calcifications is one of the major roles of mammography. The purpose of the study is to utilize some image processing techniques to enhance the detection of micro-calcifications on mammograms. Discrete wavelet transform and difference of Gaussian filter were applied to enhance mammograms and hybrid of spatial and frequency features with self-organizing map of neural network were used to estimate the efficiency of locating micro-calcifications. In the enhancement process, the raw image was enhanced by gradient enhancement, mean contrast enhancement, and discrete wavelet transform and difference of Gaussian filter. Mean, variance, direct cosine transform coefficients, and entropy- these were extracted in the assessment stage. Finally, a self-organizing map neural network with fuzzy criterion classifier was adopted to classify the regions with similar characteristics. Twenty mammograms, with different mammographic patterns and densities, were evaluated with common agreement of the breast imaging reporting and data system categories by two radiologists in simulation. The survey revealed the micro-calcification regions had a good clustering property in self-organizing map neural network index.