Multi-Task Breast Ultrasound Image Segmentation and Classification Using Convolutional Neural Network and Transformer
Click the Poster to View Full Screen, Right click to save image
Grant: Students Partnering with Faculty (SpF)
Joanna Loja
CoPIs:
Armando Mendez
College:
The Dorothy and George Hennings College of Science, Mathematics, and Technology
Major:
Information Technology
Faculty Research Advisor(s):
Kuan Huang
Abstract:
Breast Cancer in the realm of healthcare, remains a global concern and the early detection of breast tumors is crucial to improving survival rates and the quality of life for those affected. Breast ultrasound (BUS) imaging offers a non-invasive and radiation-free method to examine breast tissues. Automated BUS image segmentation and classification can help doctors identify lesions and possible abnormalities early, enabling healthcare professionals to detect breast cancer or other conditions in time for early intervention. In this research, we first conduct a comprehensive performance comparison between transformer networks and convolutional networks; secondly, we propose a novel approach by merging segmentation and classification networks, creating a multi-task network tailored explicitly for BUS image segmentation and classification; thirdly, we thoroughly investigate network performance and refine training parameters to prevent overfitting. Finally, we create a user-friendly GUI demo showing our classification and segmentation results. The results demonstrate that the ResNet-50 Multi-Task model exhibits the best overall performance for both segmentation and classification tasks.