Breast Ultrasound Image Classification Using EfficientNetV2 and Shallow Neural Network Architectures

Abstract

Health is the foundation of life. Along with the development of science and technology, and modern medicine, there are many methods to help diagnose diseases quickly and accurately detect diseases early with ultrasound images are the most popular and effective technique. This study has investigated and explored methods of predicting breast cancer based on ultrasound images classified into three categories: normal, benign, and malignant, with three deep learning techniques such as Fully Connected neural Network (FCN), Shallow Convolutional Neural Networks (CNN), and EfficientNetV2. Such techniques are evaluated on both the original data set and the dataset are applied data augmentation techniques to compare the efficiency of each algorithm and evaluate the influence of increasing data on the training phase. Experimental results show that the recent emerging CNN architecture, EfficientNetV2 provides the highest classification performance when taking the lead in all three metrics of ACC, AUC, and MCC with 0.77, 0.804, and 0.619, respectively. The results also show that data augmentation techniques can improve the efficiency of the classification algorithms on the ultrasound image dataset and are expected to combine the recent emerging CNN architectures to provide efficient disease diagnosis on ultrasound images.