Optimization of U-shaped pure transformer medical image segmentation network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Neural networks, particularly U-shaped deep neural networks, are crucial for medical image segmentation.
• Accurate lung segmentation is vital for early lung disease diagnosis and clinical decision-making.
• Existing methods face challenges with low precision in lung segmentation.

Purpose of the Study:

• To enhance the precision of lung segmentation using an optimized pure Transformer U-shaped network.
• To improve early diagnosis and clinical decision-making for lung diseases through accurate segmentation.
• To address the limitations of low precision in current segmentation techniques.

Main Methods:

• Proposed an optimized pure Transformer U-shaped segmentation network.
• Incorporated skip connections and special splicing techniques to reduce information loss during encoding.
• Enhanced information flow during the decoding process to improve segmentation accuracy.

Main Results:

• Achieved 97.86% accuracy in segmenting the "Chest Xray Masks and Labels" dataset.
• Demonstrated superior performance compared to full convolutional networks.
• Outperformed combined Transformer and convolution approaches.

Conclusions:

• The optimized Transformer U-shaped network significantly improves lung segmentation accuracy.
• This advancement supports more reliable early diagnosis and clinical decision-making for lung diseases.
• The proposed method represents a state-of-the-art approach in medical image segmentation.