Computerized Segmentation Method for Nonmasses on Breast DCE-MRI Images Using ResUNet++ with Slice Sequence Learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate segmentation of nonmasses in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is crucial for breast cancer diagnosis.
• Existing segmentation methods often struggle to effectively utilize temporal information present in multi-phase DCE-MRI.
• Developing advanced computational tools is essential for improving the accuracy and efficiency of nonmass analysis.

Purpose of the Study:

• To develop and evaluate a novel computerized segmentation method for nonmasses in breast DCE-MRI.
• To incorporate temporal information analysis using slice sequence learning and cross-phase convolution.
• To enhance the accuracy of nonmass detection and shape characterization.

Main Methods:

• A ResUNet++ architecture was employed, enhanced with slice sequence learning and cross-phase convolution.
• Temporal information was captured by creating 3D tensors from sequential ROI slice images across different MRI phases.
• A convolutional long short-term memory layer was integrated into the decoder for analyzing image sequences.

Main Results:

• The proposed method achieved an average nonmass detection accuracy of 90.5%.
• Performance metrics including Jaccard coefficient (0.563), Dice similarity coefficient (0.712), positive predictive value (0.714), and sensitivity (0.727) surpassed those of 3D U-Net, V-Net, and nnFormer.
• The method demonstrated high detection and shape accuracies, with an average of 1.91 false positives.

Conclusions:

• The developed ResUNet++ based method effectively segments nonmasses in DCE-MRI by leveraging temporal information.
• The integration of cross-phase convolution and convolutional LSTM significantly improves segmentation performance.
• This technique shows promise for aiding in the differential diagnosis of nonmasses in clinical practice.