A Binary Classification Model Based on VGG-16 for Breast Cancer Prediction

Abstract

Breast cancer is a great problem among women worldwide and constitutes nearly a quarter of all cancer cases. Early and precise detection is the first step toward saving lives and improving the living condition of patients. Even though there have been several developments in medical imaging and diagnostic technologies, the problems such as misdiagnoses, mistakes in expert interpretations, and lack of resources still remain. Traditional detection methods like mammography, ultrasound, and MRI rely heavily on expert interpretation and therefore may produce false positives and negatives, which is why the demand for automated, reliable, and alternative methods is rising. This article presents a study which leverages DL, particularly the VGG-16 model to improve the diagnosis of breast cancer from mammography images. The VGG-16 DL model that was initially trained was later fine-tuned to perform a two-class classification (Cancer and Non-Cancer) on a superior-quality dataset from Kaggle. The preprocessing steps made sure that the data was accurate and suitable, whereas, an 80-20 split of data was utilized in the training and testing phases to leave no stones unturned in the evaluation. The suggested architecture was capable of achieving quite an impressionable accuracy rate of 97.6% plus very excellent sensitivity, specificity, and precision, which were beyond most current methods for breast cancer classification. On the other hand, there are quite a handful of hurdles faced by AI deployment in healthcare, such as the quality of data, explanatory power of the model, and ethical aspects. Nonetheless, the paper focuses mainly on revealing the extraordinary power of DL to resolve the core issue of breast cancer at an extremely fundamental level.