article · Results in Engineering
Breast cancer remains a major cause of mortality among women globally, creating a continuous demand for precise and timely diagnostics. This research presents an artificial intelligence pipeline that pairs deep transfer learning with U-Net segmentation to analyse breast ultrasound scans. Using a curated dataset classified into normal, benign, and malignant cases, the study evaluated pre-trained networks alongside a segmentation model. Among the tested classification architectures, VGG19 delivered the strongest results, achieving 95.5 percent accuracy alongside superior precision and recall compared to other models. Simultaneously, the U-Net architecture accurately mapped tumour boundaries, securing an average Dice Similarity Coefficient of 85.97 percent. Together, these tools provide automated lesion localisation and classification that could help reduce diagnostic variability.
Ultrasound interpretation can vary significantly depending on the clinician examining the scan. By automating the classification and boundary detection of breast tumours, artificial intelligence can improve diagnostic precision. This supports earlier cancer detection, reduces discrepancies between medical observers, and helps clinicians make faster, more informed decisions for patient care.
The pipeline is aimed at medical imaging software developers and clinical healthcare providers seeking computer-aided diagnostic tools for ultrasound workflows. It sits at an applied and tested research stage, having been validated on a curated dataset. Practical commercial deployment will require resolving real-world barriers mentioned in the research, including computational requirements, class imbalances, and validating the software on more diverse patient populations.
AI-generated from the published abstract. Always read the original work before citing.
• This study combines deep transfer learning with U-Net segmentation to revolutionize breast cancer diagnostics. • VGG19 achieved an accuracy (95.5 %) in classifying breast ultrasound images into normal, benign, and malignant categories. • The U-Net model demonstrated high segmentation precision with an average Dice Coefficient of 85.97 %. • Transfer learning models outperformed traditional machine learning and custom CNNs in accuracy and efficiency for breast cancer imaging. • The research underscores AI’s potential to transform diagnostic workflows, enabling earlier and more precise breast cancer detection. Breast cancer remains one of the leading causes of mortality among women worldwide, highlighting the need for timely and accurate diagnostic strategies. This study investigates the integration of artificial intelligence (AI) techniques, specifically deep transfer learning for classification and U-Net for segmentation to improve breast cancer diagnosis using ultrasound imaging. A curated dataset of breast ultrasound images, categorized as normal, benign, or malignant, was used for model evaluation. Three pre-trained convolutional neural networks (CNNs), including VGG16, VGG19, and EfficientNet were implemented within a deep transfer learning framework due to their strong feature extraction capabilities. In parallel, the U-Net model, recognized for its effectiveness in medical image segmentation, was employed to delineate tumour boundaries with high spatial precision. Among the CNN models, VGG19 achieved the best performance, with the highest weighted accuracy, precision, and recall. U-Net attained an average Dice Similarity Coefficient of 85.97 %, underscoring its proficiency in segmenting tumour regions across varying lesion types. These AI-based models offer a robust diagnostic pipeline that improves lesion localization, reduces interobserver variability, and supports clinical decision-making. The approach aligns with Sustainable Development Goal (SDG) 3 by promoting early detection and better health outcomes, and SDG 9 through the adoption of innovative AI technologies in healthcare. However, limitations persist, including computational demands, class imbalance, and the lack of dataset diversity, which may affect generalizability. Addressing these challenges is essential for the safe and effective deployment of AI in real-world clinical settings.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1016/j.rineng.2025.105047
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
