A novel deep CNN model for improved breast cancer detection using ultrasound images

The biggest cause of cancer related fatalities in women globally is breast cancer. So, the early detection of this disease is beneficial to the patient since it allows for the adaptation of appropriate care. Breast ultrasonography images with complex artifacts and high noise levels make diagnosis extremely difficult for radiologists. Automatic breast cancer classification is therefore our top objective. In this article, a variety of cutting-edge deep convolutional neural networks (CNNs) are used for breast cancer classification, including AlexNet, VGG-16, and ResNet-18 etc. A novel CNN model named as BCI-Net has been proposed for classifying breast cancer. As activation function plays a very important role for the performance and training dynamics in neural network, in this model the Mish activation function is applied. For this experiment, a publicly available ultrasound images dataset named Breast Ultrasonography Images (BUSI) is used. Throughout the experiments, the proposed model has gained 98.70% average accuracy using hold-out validation; where it has acquired 97.49% overall average accuracy with a standard deviation (sd) of 1.14% after doing five fold cross-validation. In the clinical diagnosis, this analysis can be used as a secondary strategy of breast cancer detection.