A Dual-stage Deep Learning Framework for Breast Ultrasound Image Segmentation and Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer is a leading cause of death among women, making early detection critical.
• Deep Learning (DL) shows promise in enhancing medical image analysis for diagnostics.
• Ultrasound imaging is a key tool for breast mass evaluation.

Purpose of the Study:

• To apply Deep Learning techniques for segmenting and classifying breast masses in ultrasound images.
• To develop a dual-stage pipeline for improved breast cancer diagnosis.
• To evaluate the performance of different DL architectures for this task.

Main Methods:

• A modular, dual-stage DL pipeline was proposed: segmentation followed by classification.
• The pipeline flexibly integrates various backbone architectures (e.g., ResNet34, MobileNetV3-Small, EfficientNet-B0).
• An ablation study was performed to optimize model parameters.

Main Results:

• DeepLabV3+ with ResNet34 achieved the most accurate segmentation of suspicious regions.
• Lightweight classifiers MobileNetV3-Small and EfficientNet-B0 demonstrated superior classification performance.
• The approach showed promising improvements in diagnostic accuracy on two breast ultrasound datasets.

Conclusions:

• The proposed DL pipeline effectively segments and classifies breast masses using ultrasound images.
• The method has the potential to significantly enhance early breast cancer detection and diagnostic accuracy.
• Flexible integration of DL architectures allows for task-specific optimization.