Automatic Detection of Benign/Malignant Tumor in Breast Ultrasound Images using Optimal Features

Background: Breast cancer (BC) is one of the most severe diseases in women. Therefore, a premature diagnosis is necessary for timely detection and treatment execution. Clinical-level diagnosis of BC is normally performed with imaging techniques, and Ultrasound-Imaging (UI) is one of the noninvasive imaging techniques frequently executed to diagnose BC. Aims: This research aims to develop an efficient deep-learning framework to detect BC from UI with better accuracy. Methods: The executed method consists of the following stages: (i) Data collection and preprocessing, (ii) Deep-features mining with pre-trained VGG16, (iii) Image enhancement using Discrete Wavelet Transform (DWT) and Local Binary Pattern (LBP), (iv) Firefly-algorithm (FA) supported feature reduction, and (v) Feature integration and classification. Results: The proposed work is tested and executed using 1680 test images (840 benign and 840 malignant) of dimension pixels and implements a binary classifier with 5-fold cross-validation to separate the UI database into the healthy/cancer class. Conclusion: This work implemented FA-supported feature reduction. Moreover, it was found that this scheme helps to achieve a classification accuracy of 98.21% with the KNN classifier.