PLU-Net: Extraction of multiscale feature fusion

Area of Science:

• Medical Imaging
• Deep Learning
• Computer Vision

Background:

• Deep learning excels in medical image segmentation but struggles with image boundaries and details.
• Large networks often lead to suboptimal segmentation due to parameter challenges.

Purpose of the Study:

• To develop an improved deep learning model for accurate medical image segmentation, particularly focusing on boundary delineation.
• To enhance semantic information extraction and local feature representation in medical images.

Main Methods:

• Developed the PS module with atrous spatial pyramid pooling (ASPP) and SE block for broader receptive fields and detailed semantic information.
• Proposed the LG block and LS block (LG + SE) to capture local features and preserve edge information during down-sampling.
• Integrated PS and LS modules into U-Net to create the proposed PLU-Net architecture.

Main Results:

• PLU-Net demonstrated superior performance in medical semantic segmentation tasks across three benchmark datasets.
• The model achieved better results compared to existing methods while utilizing fewer parameters and FLOPs (floating-point operations per second).
• Enhanced retention of edge information and improved segmentation of image boundaries were observed.

Conclusions:

• PLU-Net offers an effective and computationally efficient solution for medical image segmentation.
• The proposed PS and LS modules significantly contribute to improved accuracy in capturing semantic and local features, respectively.
• PLU-Net represents a promising advancement for precise medical image analysis and segmentation.