Classification of breast lesions in ultrasound images using deep convolutional neural networks: transfer learning versus automatic architecture design

• DOI: 10.1007/s11517-023-02922-y
• Publication Date: 2023-09-22T02:01:27Z
• Authors: Alaa AlZoubi; Feng Lu; Yicheng Zhu; Tao Ying; Mohmmed Ahmed; Hongbo Du
• Keywords: Transfer of learning; Robustness; Contextual image classification; Bayesian Optimization

Abstract Deep convolutional neural networks (DCNNs) have demonstrated promising performance in classifying breast lesions 2D ultrasound (US) images. Exiting approaches typically use pre-trained models based on architectures designed for natural images with transfer learning. Fewer attempts been made to design customized specifically this purpose. This paper presents a comprehensive evaluation learning solutions and automatically networks, analyzing the accuracy robustness of different recognition three folds. First, we develop six DCNN (BNet, GNet, SqNet, DsNet, RsNet, IncReNet) Second, adapt Bayesian optimization method optimize CNN network (BONet) lesions. A retrospective dataset 3034 US collected from various hospitals is then used evaluation. Extensive tests show that BONet outperforms other models, exhibiting higher (83.33%), lower generalization gap (1.85%), shorter training time (66 min), less model complexity (approximately 0.5 million weight parameters). We also compare diagnostic all against by experienced radiologists. Finally, explore saliency maps explain classification decisions models. Our investigation shows can assist comprehending decisions. Graphical