Contourlet Transformed Nakagami Image-based Breast Tumor Classification Using Deep CNN

Abstract

It is still unexplored territory to automate the distinction of breast tumors using breast ultrasound (US) images. The parametric image-based approach described in this research employs convolutional neural network architecture using breast ultrasound images, to categorize and determine which breast tumors are benign and which are malignant. In this case, the contourlet transform is used to summarize the statistics of ultrasound imaging data using the Nakagami distribution as a helpful model. The proposed convolutional neural network is applied to classify breast tumors by parametric imaging produced by values of the Nakagami distribution's dispersion parameters which were calculated locally in various contourlet sub-bands. In this experiment, 100 benign fibroadenomas and 100 malignant fibroadenomas were considered from a publicly accessible dataset of 780 breast US images. The proposed technique has a 96% accuracy, 96.94% sensitivity, 95.1% specificity, 95% PPV, and 97% NPV, respectively. Furthermore, the Proposed Method's precision is superior to few recently reported findings.