Attention-based Fusion Network for Breast Cancer Segmentation and Classification Using Multi-modal Ultrasound Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer is a leading cause of cancer in women, making early detection critical for improved patient outcomes.
• Ultrasound (US) imaging is a vital, accessible tool for breast cancer screening.
• Deep learning (DL) advancements are enhancing automated analysis of medical images for computer-aided diagnosis (CAD).

Purpose of the Study:

• To develop and evaluate a novel deep learning-based model for the segmentation and classification of breast lesions using B- and SE-mode ultrasound images.
• To improve the accuracy and reliability of automated breast cancer detection through advanced image analysis.

Main Methods:

• A Multi-Modal Fusion U-Net (MMF-U-Net) was developed for lesion segmentation by integrating B- and SE-mode ultrasound data.
• A segmentation mask was used to crop lesion areas for subsequent classification.
• The encoder of the MMF-U-Net model was utilized for classifying segmented lesions as benign or malignant.

Main Results:

• The MMF-U-Net achieved strong segmentation performance with a Dice score of 78.23% and Intersection over Union (IoU) of 68.60%.
• Classification accuracy for distinguishing benign from malignant lesions reached 98.46%.
• The model demonstrated high precision (82.21%) and recall (80.58%) in segmentation tasks on clinical data.

Conclusions:

• The proposed deep learning model effectively segments breast lesions from ultrasound images.
• The method reliably classifies lesions as benign or malignant, supporting clinical decision-making.
• This approach shows significant potential for enhancing early breast cancer diagnosis via automated US image analysis.