Breast Ultrasound Classification Using Pretrained Deep Learning Models
Majd F Omari, Rasha Obeidat · 2025
Breast cancer is a major health challenge worldwide, with high incidence and mortality rates among women. Early and accurate diagnosis is crucial to improve survival rates and reduce unnecessary treatments. However, diagnostic errors, especially during early screenings such as mammography, remain a major problem, leading to delayed treatment or unnecessary treatments and interventions. This study aims to increase the accuracy of breast cancer diagnosis using ultrasound images by applying deep learning models. The dataset comprises 780 ultrasound images classified as benign, malignant, or normal. Data augmentation and class weighting techniques were used to address the imbalance in the dataset. After evaluating multiple pre-trained deep learning models, DenseNet121 emerged as the best-performing model, achieving an accuracy of 91.45 %. DenseNet121 demonstrated superior feature sharing, fewer parameters, stable training, and efficient feature reuse, making it particularly effective for breast cancer diagnosis from ultrasound images. This study highlights the efficacy of deep learning models, especially DenseNet121, to improve the accuracy of a breast cancer diagnosis from ultrasound images, contributing to improved patient outcomes and reduced diagnostic errors.