DNUNet: A lightweight adaptive medical image segmentation network based on dual-path multilevel interactive...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep learning excels in medical image segmentation, crucial for diagnosis and treatment.
• Clinical deployment demands efficient, lightweight models for portable devices and real-time processing.
• Balancing segmentation accuracy with computational cost is a key challenge in medical AI.

Purpose of the Study:

• To propose DNUNet, an efficient and lightweight deep learning model for medical image segmentation.
• To enhance feature extraction and fusion while minimizing computational and memory overhead.
• To achieve a balance between model performance and resource consumption for clinical applications.

Main Methods:

• Utilized large kernel convolution, a dual-path multilevel structure, and feature sparsification.
• Designed a dual-path multilevel interactive convolutional module for increased network depth with fewer parameters.
• Introduced an adaptive norm sparse fusion module as an alternative to traditional skip connections for efficient feature fusion.

Main Results:

• DNUNet achieves high-precision medical image segmentation with significantly reduced computational and memory requirements.
• The model demonstrates a good balance between lightweight architecture and performance across multiple datasets.
• Outperformed various state-of-the-art (SOTA) methods in segmentation tasks.

Conclusions:

• DNUNet offers a promising solution for efficient and accurate medical image segmentation.
• Its high efficiency and low resource consumption enable potential real-time deployment in clinical scenarios, including portable devices.
• The proposed model effectively addresses the challenge of balancing performance and computational cost in medical AI.