Performance evaluation of modified Gaussian mixture model algorithm for image segmentation of mammogram

Computer-assisted diagnosis (CAD) in radiological practices has been achieving newer heights due to advances in image processing techniques and computing facilities. Detection of microcalcification in full-field digital mammography using image segmentation method has been studied in this paper. The proposed technique comprises use of adaptive median filter for removal of speckle noise and the use of modified Gaussian mixture model for image segmentation. In finite mixture model, the spatial relationship between the neighbourhood pixels is not considered but the proposed modified Gaussian mixture model removes the above mentioned limitations by considering the conditional probability. Inclusion of conditional probability incorporates the spatial information of neighbourhood pixels in the model. For simulation and validation purpose, mini-MIAS and mini-DDSM datasets of mammography images are considered. Qualitative and quantitative analyses have been carried out using different performance metrics to validate the performance of the proposed modified Gaussian mixture model for segmentation.