UCapsNet: A Two-Stage Deep Learning Model Using U-Net and Capsule Network for Breast Cancer Segmentation and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer is a major global health concern for women.
• Ultrasound imaging aids tumor detection but often lacks detail for accurate diagnosis.
• Traditional U-Net models struggle with image quality, impacting early detection rates.

Purpose of the Study:

• To develop an advanced model for improved breast cancer detection in ultrasound images.
• To overcome the limitations of conventional U-Net models in identifying subtle tumor features.
• To enhance both segmentation and classification accuracy for more reliable diagnoses.

Main Methods:

• An enhanced U-Net model integrated with a Capsule Network (UCapsNet) was proposed.
• The model utilizes higher filter counts and skip connections for improved segmentation.
• A two-stage process involves U-Net segmentation followed by Capsule Network classification.

Main Results:

• UCapsNet achieved high performance metrics on the Breast Ultrasound Image (BUSI) dataset.
• Precision, recall, and accuracy rates were recorded at 98.12%, 99.52%, and 99.22%, respectively.
• The model outperformed established pre-trained models like VGG-19, DenseNet, and ResNet-50.

Conclusions:

• UCapsNet effectively combines segmentation and classification for enhanced diagnostic precision.
• The model addresses key weaknesses in existing breast cancer detection methods.
• Findings support UCapsNet's reliability for practical clinical applications in early tumor detection.