Multi-scale contextual semantic enhancement network for 3D medical image segmentation.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Computational pathology

Background:

• Accurate medical image segmentation is vital for disease diagnosis and treatment planning.
• Existing convolutional neural network methods struggle with lesion scale variations, blurred boundaries, and class imbalance.

Purpose of the Study:

• To develop a novel segmentation framework, the 3D multi-scale contextual semantic enhancement network (3D MCSE-Net), to overcome current limitations in medical image segmentation.
• To enhance the accuracy and efficiency of tumor segmentation in medical imaging.

Main Methods:

• The 3D MCSE-Net incorporates a multi-scale context pyramid fusion module (MCPFM) to handle scale variations.
• A triple feature adaptive enhancement module (TFAEM) is used to refine lesion boundaries.
• An asymmetric class correction loss (ACCL) function addresses class imbalance issues.

Main Results:

• The 3D MCSE-Net demonstrated superior performance on nasopharyngeal cancer tumor segmentation (NPCTS), liver tumor segmentation (LiTS), and 3Dircadb datasets.
• Experimental results confirmed the effectiveness and generalizability of the proposed modules and the overall framework.
• The integrated components showed mutually reinforcing properties, leading to improved segmentation accuracy.

Conclusions:

• The 3D MCSE-Net effectively addresses challenges in medical image segmentation, including scale variation, blurred boundaries, and class imbalance.
• The proposed method significantly improves tumor segmentation accuracy, aiding clinical diagnosis and treatment planning.
• This framework offers a promising advancement for automated medical image analysis.