Evaluating Optimizer Performance in Convolutional Neural Networks for Breast Microcalcification Classification: A Comparative Investigation

Deep neural network training is significantly impacted by optimizers, which have an impact on the network's generalization ability, convergence speed, and overall efficiency of the model. Choosing the best optimizer is critical in the context of breast microcalcification classification, where accurate and precise diagnosis is essential for successful treatment. In this work, Convolutional Neural Networks (CNNs) were used to examine the effects of different optimizers on the categorization of breast microcalcifications. Specifically, the Chinese Mammography Database (CMMD) was used to assess the effectiveness of well-known optimizers including Adam, RMSprop, Adadelta, and Adagrad in terms of optimizing CNN architectures for accurate microcalcification classification. In this work, a dataset of annotated mammograms was used to train multiple CNN models, each using a different optimizer. The comprehensive experiments were carried out to assess these optimizers' performance in terms of adaptability, convergence speed, and classification accuracy. The study's conclusions offerred insightful information about the choice of optimizers for CNN optimization in the context of medical image analysis, especially for the microcalcification classification-based early diagnosis of breast cancer.