Breast tumor segmentation via deep correlation analysis of multi-sequence MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate breast tumor segmentation from MRI is vital for breast cancer diagnosis and characterization.
• Existing methods struggle to model complex interrelationships within multi-sequence MRI data.

Purpose of the Study:

• To develop a hybrid deep network framework for enhanced breast tumor segmentation using multi-sequence MRI.
• To effectively integrate and exploit spatial-temporal features across different MRI sequences.

Main Methods:

• A hybrid deep network with three modules: a multi-sequence encoder, a multi-scale feature embedding module, and a decoder.
• Utilized a sequence-awareness and temporal-awareness method for fusing spatial-temporal features.
• Employed a densely connected architecture in the encoder for individual MRI sequence processing.

Main Results:

• The proposed method achieved Dice Similarity Coefficient (DSC) of 80.57%, Intersection over Union (IoU) of 74.08%, and Positive Predictive Value (PPV) of 84.74%.
• Demonstrated notable improvements in segmentation performance compared to existing methods.
• Successfully learned interrelationships inherent in multi-sequence MRI data.

Conclusions:

• The hybrid deep network framework offers a promising approach for precise breast tumor segmentation.
• The method's ability to leverage multi-sequence MRI data leads to superior segmentation outcomes.
• This advancement can aid in more accurate breast cancer diagnosis and treatment planning.