Breast Cancer Detection Using ResNet with Hyperparameter Tuning
Jiatai Mu · 2023
There are more than 2 million new cases of breast cancer each year, which is one of the leading causes of death worldwide. Early diagnosis improves cure rates and reduces mortality and misdiagnosis. However, a common way to diagnose breast cancer is based on mammographic images, which may misdiagnose. In this paper, a convolutional neural network with a pre-trained Residual Neural Network (ResNet) is proposed to diagnose breast cancer based on histology images. The proposed model employs ResNet to extract features from the images, then classify them into benign or malignant. Then a hyperparameter tuning process using the hyperband algorithm is applied to find the optimal hyperparameters in the model. To verify the effectiveness of the proposed method, the default model is compared with the models from tuning results using the histopathological images from other datasets. The comparison result is based on the AUC, ROC, validation accuracy, and validation loss. The experimental results demonstrate that the best-modified model has a 99.63 percent of validation accuracy. The analyses illustrate that the modified ResNet model performs better on detect breast cancer based on histopathological images than the default model.