Breast cancer remains a major global health issue, being the most commonly diagnosed cancer among women worldwide and also affecting a significant number of men. Despite advancements in screening techniques such as mammography and ultrasound, there is a critical need for more precise diagnostic tools to enhance early detection and treatment. Recent developments in machine learning, particularly deep learning, have shown promising potential to improve detection accuracy by effectively analyzing complex patterns in medical imaging. However, developing effective deep learning models tailored to breast cancer data presents substantial challenges. These challenges include processing extensive datasets of breast cancer images, complex model training, and selecting optimal parameters that improve detection accuracy without compromising generalizability across different scenarios and imaging technologies. This paper proposes an innovative approach utilizing deep learning to analyze the Ki-67 protein index from biopsy samples, a crucial marker of cell proliferation in breast cancer diagnostics. By applying advanced neural architectures such as DeepLabv3+ with MobileNet-v2, Xception, DenseNet-121, U-Net, and the Fully Convolutional Regression Network, our method focuses on distinguishing between Ki-67 positive and negative tumor cells and detecting tumor-infiltrating lymphocytes with high precision. These models were rigorously evaluated against the SHIDC-B-Ki-67 dataset, achieving not only high accuracy, reaching up to 98.8%, but also significant reductions in processing times, down to just 13 s, which is crucial for timely clinical decision-making. Our results contribute to integrating artificial intelligence with conventional diagnostic methods, establishing benchmarks for the accuracy and efficiency of breast cancer detection and paving the way for future research in automated medical image analysis.