Precision Enhanced Breast Cancer Prediction Using Deep Learning Models

Breast cancer is still a major worldwide health con- cern, which emphasises how important it is to have accurate and trustworthy diagnostic techniques. Our goal is to improve breast cancer identification in this scenario by utilizing mammography pictures from the RSNA Breast Cancer dataset. Leveraging the power of deep learning, our methodology integrates the SEResNet and ConvNeXtV2 models, simultaneously exploring and comparing various methods. To enhance the performance of our models, we incorporated changes in activation functions by replacing ReLU with GELU, an alternative activation function. Furthermore, we improved downsampling by introducing sepa- rate downsampling layers in the model architecture, optimizing spatial downsampling, and employing 3×3 convolutions with a stride of 2 at the start of each stage and 1×1 convolutions with a stride of 2 at the shortcut connection. The results achieved are highly promising, with a remarkably low loss of 0.0445 and impressive performance metrics. The F1 score stands impressively high at 0.96, complemented by a recall score of 0.95 and a precision of 0.97, showcasing exceptional performance, The accuracy score is also impressive at 92.5. These findings underscore the pivotal role of precision in elevating breast cancer detection accuracy, providing a path for more accurate diagnoses and ultimately improving patient outcomes. These enhancements significantly contributed to the overall performance of our breast cancer detection models, reinforcing the importance of precision in healthcare. This research strengthens the diagnostic tools available in the ongoing fight against breast cancer.