nnUNet-based Multi-modality Breast MRI Segmentation and Tissue-Delineating Phantom for Robotic Tumor Surgery Planning

Area of Science:

• Medical imaging analysis
• Artificial intelligence in surgery
• Biomedical engineering

Background:

• Accurate segmentation of breast tissues in MRI is vital for diagnosing breast masses.
• Existing methods require robust algorithmic and phantomic foundations for surgical planning.

Purpose of the Study:

• To develop an advanced medical image segmentation architecture for breast MRI.
• To propose a novel polyvinyl alcohol cryogel (PVA-C) breast phantom for surgical planning and navigation.

Main Methods:

• Utilized a dual neural network architecture based on nnU-Net for segmenting breast MRI data.
• Employed a two-stage labeling process: single-class for breast region, then three-class for fatty, fibroglandular (FGT), and tumorous tissues.
• Developed an automated segmentation approach for the PVA-C breast phantom.

Main Results:

• Achieved a Dice Similarity Coefficient (DSC) of 0.95 for breast region segmentation.
• The second network demonstrated DSCs of 0.95 (fat), 0.83 (FGT), and 0.41 (tumor).
• The PVA-C phantom facilitates planning and navigation experiments for robotic breast surgery.

Conclusions:

• The developed deep learning algorithm and breast phantom provide a foundation for image-guided robotic breast surgery.
• This approach integrates preoperative MRI and intraoperative ultrasound for enhanced surgical planning and navigation.
• The technology aims to improve surgical accuracy, patient outcomes, and cosmetic results in breast cancer treatment.