Breast tissue density classification based on gravitational search algorithm and deep learning: a novel approach

This work presents the automatic classification of mammographic breast tissue density as it plays a crucial role in morphological analysis for abnormality detection. The proposed work consists of pre-processing, mammogram enhancement, and classification. Image pre-processing is used to extract the foreground image from the original image. Image extraction is a well known optimization problem. In order to capture the aforementioned problem, gravitation search algorithm is adopted with the consideration of kapur’s entropy as a fitness function. Afterwards, mammogram enhancement and noise removal is performed by utilizing unsharp masking and Anisotrophic filtering technique. Finally, VGG16 and InceptionV3 were utilised to train the convolution neural network (CNN) classification model using deep transfer learning. For a four class classification of breast tissue density, the suggested method is evaluated on the dataset from Digital Image Database for Screening Mammography (DDSM). It obtains classification accuracy of 97.98% for InceptionV3 based model and 91.92% for VGG16 based model on DDSM dataset. © 2022, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.