Advancements in Breast Cancer Detection: A Comprehensive Review of Deep Learning-Based Classification and Segmentation Techniques Using Ultrasound Imaging
Rahul Singh, Sheifali Gupta, Gotte Ranjith Kumar · 2025
Breast cancer remains one of the most prevalent causes of mortality among women worldwide, underscoring the critical importance of early detection and accurate diagnosis. Ultrasound imaging has gained prominence due to its accessibility, cost-effectiveness, and safety, especially for younger women and those with dense breast tissue. However, inherent challenges such as low spatial resolution, speckle noise, and variability in image quality pose significant hurdles to precise analysis. Recent advancements in deep learning techniques have revolutionized breast cancer detection, particularly in the classification and segmentation of ultrasound images. This review analyzes state-of-the-art methods utilizing existing available datasets like BUSI, OASBUD, and KAIMRC. Classification models have demonstrated impressive performance, with Himel et al. achieving 99.7% accuracy, 99.71% F1-score, and an AUC of 99.9% using ensemble deep learning approaches. Similarly, segmentation models have achieved remarkable outcomes, with Hekal et al. reporting a Dice Similarity Coefficient (DSC) of 92.92% and an Intersection over Union (IoU) of 87.39%. Integrated models combining classification and segmentation have further improved detection capabilities; for instance, Bobowicz et al. achieved 93.7% classification accuracy and an 87% DSC for segmentation. While these advancements are promising, challenges such as limited dataset diversity, variability in ultrasound imaging, and generalizability remain. Future research should explore hybrid models, transfer learning, and larger, multi-institutional datasets to enhance segmentation precision and classification reliability, paving the way for robust clinical applications.