A novel risk prediction based on adaptive adjustment VGG16 for breast cancer
Early diagnosis of breast cancer can increase the chance of cancer cure,thus it is important to improve the accuracy of early diagnosis.In this paper,combined with the deep learning framework Pytorch,a deep learning-based adaptive adjustment VGG16 model(AD-VGG16)is proposed to predict the benign and malignant classification of breast cancer.The overfitting and local minimum problems are optimized by adaptive adjustment of learning rate and equalization of positive and negative samples.The generalization a-bility of the network is improved by cross validation.The first fully connected layer of VGG16 network is modified to adapt to input samples of arbitrary size.Three models,including VGG16,ResNet18 and AD-VGG16,are used to predict the normal,benign and malignant classification of DDSM dataset.Experiment data shows that AD-VGG16 network has better classification accuracy than VGG16 and ResNet18 network,and its final test classification accuracy is 94.3％.
deep learningbreast cancerVGG16ResNet18DDSM datasets
万方数据
维普
