An Efficient DenseNet201 Pre-trained Model for Early Prediction of Breast Cancer
Gurpreet Singh, Kalpna Guleria, Shagun Sharma · 2023
Breast cancer detection is one of the biggest challenges in the entire health care system. As per the report by WHO, in 2020, breast cancer resulted in 6,85,000 deaths globally. It has been identified that this type of cancer generally occurs in women, whereas the ratio of occurrence in men is 0.5–1% only. In this type of cancer, the abnormal cells start to grow and result in the development of a tumour. The increased growth rate of breast cancer cases can be reduced if detected at its early stages. This work aims to investigate the use of the DenseNet201 deep learning model in the context of breast cancer early prediction via the analysis of mammography images. The dataset used in this study has been collected from Kaggle, which contains both benign and malignant cases of breast cancer. The images have been preprocessed to reduce the abnormalities and result in error-free outcomes. The central aspect of the proposed methodology is based on DenseNet201 architecture, which is a convolutional neural network that has been pre-trained on the ImageNet dataset. During the course of the investigation, the implementation of comprehensive training and validation procedures, using rigorous cross-validation methods to guarantee the generalizability of the model has been done. The evaluation metrics included in this study are training accuracy, testing accuracy, training loss, and testing loss. This model has been implemented for different epoch values and the results have shown that for training the model the highest accuracy of 88.50% has been achieved at epoch 70, whereas the testing accuracy of 96.88% has been identified at epoch 70. The lowest training and validation loss have been also identified at epoch 70 as 0.323, and 0.131, respectively.