Consistent Learning-Based Breast Tumor Segmentation and Its Application in Sentinel Lymph Node Metastasis Prediction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate breast tumor segmentation is vital for staging lymph nodes in breast cancer patients.
• Current segmentation methods struggle with variations in tumor size, image quality, and noisy annotations.
• Developing robust segmentation techniques is crucial for improving diagnostic accuracy.

Purpose of the Study:

• To develop a Multi-scale RepVGG-based Segmentation Network (MPSegNet) for segmenting breast tumors from MR images.
• To implement a consistent learning framework to mitigate the impact of noisy labels on segmentation.
• To analyze the relationship between segmentation performance and the prediction of sentinel lymph node (SLN) metastasis.

Main Methods:

• Developed MPSegNet, a novel deep learning model incorporating a multi-scale RepVGG backbone.
• Constructed a consistent learning framework ensuring identical segmentation predictions from different views of the same tumor.
• Evaluated segmentation accuracy and its correlation with SLN metastasis prediction performance.

Main Results:

• MPSegNet demonstrated superior performance compared to existing state-of-the-art methods.
• Consistent learning significantly enhanced breast tumor segmentation accuracy.
• Optimal segmentation did not directly correlate with the best SLN metastasis prediction, indicating a complex relationship.

Conclusions:

• The proposed MPSegNet with consistent learning offers improved breast tumor segmentation in MR images.
• The study highlights the importance of investigating the intricate relationship between tumor segmentation and metastasis prediction for precise patient care.
• This work has potential significance for enhancing the medical care of breast cancer patients through accurate segmentation and predictive analysis.