An Efficient Deep Convolutional Network Model using Mask İmages for Multiclass Classification of Breast Cancer Ultrasound İmages

Abstract

Breast cancer begins in the breast tissue when mutated cells grow out of control and eventually form a tumor. One of the most common causes of death among women worldwide is breast cancer. Early diagnosis and treatment can increase the likelihood of cancer prevention and recovery. Breast ultrasound analysis performed by medical professionals requires high competence in interpreting images, is time-consuming, and creates a negative situation in terms of the treatment process. Artificial intelligence methods have shown great success in the development of medical diagnosis and diagnostic models. When combined with artificial intelligence techniques, breast ultrasound images can produce good results in the detection and classification of breast cancer. This study focused on multiclass classification of breast cancer ultrasound images collected via ultrasound scanning via deep learning methods. In the first stage of the model, image preprocessing was performed to reduce the negative effects of poor contrast, unclear target areas, image transients, and unbalanced image classes. In the second stage, a new dataset was created by using mask images. The third stage consists of training two different datasets on a block basis with the fine-tuned MobiLenetV3Large, MobiLenetV3Small, MobiLenetV2, ResNet101 and EfficientNetV2M deep learning models and, finally, the classification stage. As a result of the experiments conducted on the breast ultrasound image datasets prepared under different conditions, the proposed model achieved 100% accuracy and F1 score for the training data, a 99.15% accuracy and a 99.21% F1 score for the test data, and better performance efficiency than the other models did.