A robust and efficient AI assistant for breast tumor segmentation from DCE-MRI via a spatial-temporal framework

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Dynamic contrast-enhanced MRI (DCE-MRI) is vital for breast tumor screening, diagnosis, and follow-up due to its high sensitivity.
• Accurate segmentation of breast tumors in DCE-MRI is critical for clinical decision-making, providing essential location and shape information.
• Current manual segmentation is time-consuming and can be a bottleneck in the diagnostic workflow.

Purpose of the Study:

• To develop an artificial intelligence (AI) assistant for automated breast tumor segmentation.
• To capture dynamic changes in multi-phase DCE-MRI using a spatial-temporal framework.
• To enhance the efficiency and accuracy of breast tumor segmentation in clinical practice.

Main Methods:

• Development of a spatial-temporal AI framework for automated tumor segmentation.
• Utilizing multi-phase DCE-MRI data to capture dynamic changes.
• Validation on a large-scale dataset from seven medical centers to ensure robustness.

Main Results:

• The AI assistant demonstrated robustness, handling MR data with varying phase numbers and imaging intervals.
• Achieved significant efficiency, reducing manual annotation time by a factor of 20.
• Maintained accuracy comparable to that of expert physicians.

Conclusions:

• The developed AI assistant provides accurate and efficient automated breast tumor segmentation from DCE-MRI.
• This AI tool is robust across diverse datasets and significantly reduces manual annotation workload.
• The AI assistant serves as a foundational step for AI-assisted breast cancer diagnosis systems, promoting clinical AI adoption.