A multi-scale attention-based Swin transformer model for medical images segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate medical image segmentation is vital for disease diagnosis and clinical decision-making.
• Current AI models often suffer from high computational demands and suboptimal accuracy.
• A need exists for efficient and accurate AI solutions in medical image analysis.

Purpose of the Study:

• To develop an optimized AI architecture for medical image segmentation.
• To improve segmentation accuracy while reducing model complexity and computational cost.
• To enhance the diagnostic process through AI-assisted image analysis.

Main Methods:

• A novel transformer-based architecture utilizing Swin transformers as an encoder for deep feature extraction.
• Implementation of a dynamic feature fusion block (DFFB) in the decoder for multi-scale feature enhancement.
• Integration of a dynamic attention enhancement block to refine features using spatial and channel attention.

Main Results:

• The proposed model achieved high segmentation performance across three medical datasets (GlaS, PH2, Kvasir-SEG).
• Achieved a mean intersection over union (mIoU) of 0.9125 and Dice score of 0.9542 on GlaS.
• Demonstrated superior performance compared to existing methods, indicating effectiveness in segmenting complex regions.

Conclusions:

• The developed AI model offers a promising solution for accurate and efficient medical image segmentation.
• The optimized architecture balances high accuracy with reduced parameters and computational load.
• This approach has the potential to significantly aid physicians in disease diagnosis and examination.