Manuscript received June 13, 2025; revised July 15, 2025; accepted August 6, 2025; published November 25, 2025.

Abstract—This study presents a deep learning-based approach to enhance the accuracy of breast cancer diagnosis through ultrasound image segmentation. Several segmentation models were evaluated, including the newly developed Breast Cancer Analysis and Recognition Enhancement-Feature Extraction Zone Network (BCARE-FEZNET), alongside U-Net, Fully Convolutional Network (FCN), ResUnet, SegNet, and Mask Region Convolutional Neural Network (Mask R-CNN). The evaluation criteria included accuracy, loss, and an in-depth analysis using metrics such as Confusion Matrix, Intersection over Union (IoU), Dice Similarity Coefficient (DSC), and Area Under Curve (AUC). The results demonstrate that BCARE-FEZNET outperforms the other models, achieving approximately 92% accuracy, high IoU and Dice values (0.85 and 0.87 respectively), with moderate AUC performance (0.56), indicating strength in localization but limited discriminative capacity for classification thresholds. While ResUnet delivers the highest AUC (0.76), it suffers from significant imbalances between object detection and false positives. Other models, such as U-Net, FCN, and Mask R-CNN, exhibit AUC values close to random guessing, while SegNet encounters instability during training. Overall, BCARE-FEZNET provides the most stable and reliable performance, proving to be the optimal model for ultrasound image segmentation in breast cancer diagnosis.

Keywords—segmentation, breast cancer diagnosis, ultrasound images, Breast Cancer Analysis and Recognition Enhancement-Feature Extraction Zone Network (BCARE-FEZNET), model evaluation

Cite: Agus P. Windarto, Iwan Purnama, and Rahmadani Pane, "A Deep Learning Approach to Enhance Breast Cancer Diagnosis Accuracy through Ultrasound Image Segmentation," Journal of Image and Graphics, Vol. 13, No. 6, pp. 579-589, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.