MLFA-UNet: A multi-level feature assembly UNet for medical image segmentation

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Accurate medical image segmentation is vital for diagnosis and treatment.
• Fully convolutional networks (FCNs), particularly U-Net architectures, are prominent for medical image segmentation.
• Existing methods face challenges in capturing both local details and contextual information for precise lesion identification.

Purpose of the Study:

• Introduce MLFA-UNet, an innovative U-Net based framework for advanced medical image segmentation.
• Enhance segmentation robustness and precision by integrating novel attention mechanisms.
• Improve the accurate identification of lesions across various medical imaging modalities.

Main Methods:

• Developed MLFA-UNet, a U-shaped architecture incorporating Multi-Level Feature Assembly (MLFA) and Multi-Scale Information Attention (MSIA) modules.
• Integrated a pixel-vanishing (PV) attention mechanism to augment feature diversity and receptive field.
• Employed MLFA in encoder/decoder for local information extraction and MSIA in the bottleneck for contextual understanding.

Main Results:

• MLFA-UNet demonstrated superior performance over state-of-the-art algorithms on diverse datasets.
• Achieved high Dice coefficients: 91.42% (MICCAI 2017 Red Lesion), 82.43% (ISIC 2017), 90.8% (PH2), and 88.68% (CVC-ClinicalDB).
• Evaluated on wireless capsule endoscopy, colonoscopy, and dermoscopic images, showcasing versatility.

Conclusions:

• MLFA-UNet effectively captures detailed local and broad contextual information for enhanced segmentation.
• The proposed architecture offers improved accuracy and resilience in lesion identification.
• MLFA-UNet represents a significant advancement in medical image segmentation technology.