A streamlined U-Net convolution network for medical image processing.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical image segmentation is vital for accurate diagnosis.
• Traditional U-Net models struggle with global features and multi-scale data.
• There is a need for efficient segmentation models with reduced computational demands.

Purpose of the Study:

• To introduce the LUNeXt model for improved medical image segmentation.
• To address limitations of U-Net in feature extraction and multi-scale information.
• To develop a model balancing performance with computational efficiency.

Main Methods:

• Developed the LUNeXt model, combining Vision Transformers (ViT) with novel convolution blocks.
• Utilized depthwise separable convolutions for efficient global feature extraction.
• Conducted experiments on four diverse medical image datasets.

Main Results:

• LUNeXt achieved competitive segmentation performance.
• The model significantly reduced parameters and floating-point operations (FLOPs) compared to U-Net.
• Explainable AI techniques visualized segmentation results, confirming efficacy.

Conclusions:

• LUNeXt enables efficient medical image segmentation on standard hardware.
• The model lowers the learning curve for advanced segmentation techniques.
• LUNeXt offers a balanced approach for accurate pathological feature extraction.