Early Detection of Breast Cancer using An Improved Deep-Learning Model
Muhamamd Yousif, Hamza Alrababah, Muhamamd Atif, Munir Ahmad, Muhamamd Asghar Khan · 2023
Deep learning algorithms are now being employed to undertake the inquiry, and preliminary results have been assessed as encouraging. Because they rely on comparably small breast pathology imaging datasets and an overoptimistic version of the softmax-cross entropy loss, popular convolutional neural networks (CNNs) and their derivatives are prone to overfitting in their classifiers. This is because their classifiers are prone to overfitting. Furthermore, these CNNs and their variants employ an unduly pessimistic application of the softmax-cross entropy loss. To perform computations, these CNNs use a modified version of the overconfident softmax-cross entropy loss approach. This is the outcome of the union of these two factors. AlexNet, VGGNet, and GoogleNet are a few examples of these networks. This is because both occurrences result in the loss of the overconfident softmax-cross entropy. This is the root of the problem. We present the AlexNet-BC model to alleviate overfitting and improve classification accuracy. It is a cutting-edge diagnostic paradigm for breast illness. This allows us to ensure the accuracy of our forecasts. This allows us to penalize overly pessimistic low-entropy output distributions while also ensuring that our forecasts appropriately represent uniform distributions. It also allows us to ensure the consistency of our output distributions. The proposed method will next be tested using datasets provided by BreaKHis, IDC, and UCSB in comparison to existing methodologies considered state-of-the-art. This will be accomplished by contrasting the proposed method with other strategies considered cutting-edge. The studies' findings show that the proposed approach outperforms what are now considered cutting-edge technologies across a wide range of zoom levels. Because of the high level of consistency and generalizability that it delivers, it has the potential to be a valuable tool in histopathological clinical computer-aided diagnosis systems. The worth of a possibility may be seen just by the fact that it is a possibility.