A novel approach for automatic tumor detection and localization in mammography images via mixture of factor analyzers based on co-clustering

This paper proposes a novel model-based method called the Extension of Mixture of Factor Analyzers for CoClustering of Images (EMFACCI) to segment grayscale mammography images into a number of blocks, one of which contains the tumor. After image preprocessing and enhancement, the EMFACCI model first determines the optimal number of row clusters and column clusters by reducing the local dimension. Using the optimal number of blocks, the proposed model changes the location of columns and rows by clustering them simultaneously until it identifies the block containing the tumor. Finally, the detected block is binarized using Fuzzy C-Mean clustering and located on the input image by the proposed model, while other blocks are removed. The performance of the proposed method is evaluated on images taken from the MIAS and DDSM datasets. The results demonstrate the effectiveness of the proposed method based on sensitivity, specificity, dice, jaccard similarity index (JSI), false-negative rate (FNR), false-positive rate (FPR), and accuracy. Our proposed method performs well in detecting tumors of different sizes and is accurate in dealing with complex and high-dimensional images. Additionally, it outperforms existing detection methods.