Deep learning network for medical volume data segmentation based on multi axial plane fusion.

Area of Science:

• Medical Imaging
• Deep Learning
• Computer Vision

Background:

• High-dimensional medical data (e.g., CT scans) offer precise anatomical information.
• Deep learning models face computational and memory challenges with high-dimensional data.
• Dimensionality reduction often compromises medical image segmentation performance.

Purpose of the Study:

• To develop an efficient deep learning segmentation network for high-dimensional medical volume data.
• To address the computational and memory constraints of existing methods.
• To improve segmentation accuracy without performance degradation.

Main Methods:

• Proposed a 2D deep learning segmentation network utilizing multi-pinacoidal plane fusion.
• The approach integrates global information across different input layers.
• Designed for compatibility with various backbone networks.

Main Results:

• The multi-pinacoidal plane fusion approach demonstrated effectiveness across different backbone networks.
• DeepUnet achieved a Dice coefficient of 0.883 and a Positive Predictive Value (PPV) of 0.982.
• The method showed significant quantitative and qualitative improvements in segmentation.

Conclusions:

• The proposed network with multi-pinacoidal plane fusion enhances medical image segmentation.
• Achieved superior quantitative and qualitative results compared to standard approaches.
• The method offers a viable solution for processing high-dimensional medical imaging data.