Empowering Early Diagnosis: Optimizers Revolutionizing Breast Cancer Detection with DenseNet

Breast cancer, a highly lethal disease with a significant global impact, requires early detection to reduce its mortality rate. The emergence of pre-trained Deep Learning (DL) algorithms, commonly utilized for object detection, image classification, and image segmentation, has advanced early breast cancer detection. Several researchers have explored early breast cancer detection using pre-trained DL algorithms in conjunction with classifiers or other existing methods. However, existing research lacks a thorough examination of optimizers and their potential influence on the behaviour of these pre-trained algorithms. This study aims to comprehensively investigate the behaviour of the pre-trained DenseNet201 model under various optimizer and learning rate configurations. The analysis reveals that the DenseNet model achieved superior performance (100% accuracy) with the SGD (Stochastic Gradient Descent), and RMSProp optimizers at a learning rate of 10^-2 and 10^-3, respectively. However, at a learning rate of 10^-4, the Adam optimizer demonstrated the highest accuracy of 99%.