Intelligent Breast Cancer Screening using Deep Learning Models
K. Manikandan, E Pavithra, M.Nisha Angeline, A. Mummoorthy, Mohaideen.A, S.Vinayagan · 2025
A significant proportion of cancer deaths are attributed to breast cancer and it point out the importance of early detection for improved chances of survival. Traditional diagnostic methods such as biopsies and imaging, often require extensive manual interpretation, leading to inconsistencies. Advances in artificial intelligence offer precise, automated, and interpretable diagnostic tools to enhance breast cancer detection. This framework integrates deep learning techniques, beginning with Zero-Shot Classification using the CLIP model, allowing initial categorization without extensive dataset-specific training. It is followed by Few-Shot Classification with a fine-tuned ResNet50 model, improving classification accuracy even with limited labeled data. Heatmaps generated through Grad-CAM highlight crucial areas in tissue samples, aiding interpretability for healthcare professionals.Performance is evaluated using precision, recall, F1-score, and accuracy, ensuring reliability. Additionally, GPT-Neo generates detailed BI-RADS-compliant reports, facilitating effective communication between radiologists and clinicians. Unlike traditional manual reporting, this automated approach enhances efficiency, consistency, and scalability while reducing human error. The framework’s ability to learn from minimal labeled data makes it highly adaptable and cost-effective for widespread clinical implementation, providing an advanced AI-driven solution for breast cancer diagnosis with improved accuracy and interpretability using Gradecam++ tool.