Breast cancer ranks as a significant health threat to women globally, making timely and accurate diagnosis critical. In response to the delays experienced in traditional laboratory analyses, a novel approach has been suggested leveraging the Biglycan protein as a biomarker for breast cancer detection. This method employs image data of the biomarker, which undergoes initial preprocessing before being analyzed through a custom Convolutional Neural Network (CNN) architecture. This innovative approach is further evaluated by comparing its performance against established pretrained models, including ResNet, MobileNet, and EfficientNet. The comparison reveals that proposed model leads in accuracy, achieving a 91.68% success rate in detecting breast cancer. Remarkably, EfficientNet closely follows, registering an 89.57% accuracy rate. This finding underscores the potential of using the Biglycan protein as a reliable biomarker in the development of advanced AI-driven diagnostics of breast cancer.