RiIG Modeled Curvelet Coefficient Image-based Breast Tumor Classification Using deep CNN
Shahriar Mahmud Kabir, Mohammed Imamul Hassan Bhuiyan · 2022
Deep learning-based automatic classification of breast tumors using ultrasound (US) B-Mode images is still a challenging research area. A deep convolutional neural network employing parametric imaging is presented to categorize benign and malignant breast tumors from ultrasound B-Mode images. The Rician Inverse Gaussian (RiIG) distribution is used to represent the statistics of breast US imaging data in the domain of the Curvelet Transform. In different Curvelet sub-bands, locally estimated values of the RiIG distribution’s dispersion parameters produce parametric images. A publicly accessible dataset of 780 breast ultrasound images along with their ground truth images is used for classification with a specially designed convolutional neural network. The proposed technique yields 97.1%, 96.6%, 97.3%, 94.3%, and 98.4% accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), respectively, higher than those of several recently reported techniques.