Attention-guided duplex adversarial U-net for pancreatic segmentation from computed tomography images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate organ segmentation in computed tomography (CT) images is vital for early disease diagnosis and treatment planning.
• Pancreas segmentation presents significant challenges due to the organ's small size and anatomical variability.

Purpose of the Study:

• To develop an effective method for pancreas segmentation in CT images.
• To address the difficulties associated with pancreas segmentation, including its small volume and shape variations.

Main Methods:

• Proposed an attention-guided duplex adversarial U-Net (ADAU-Net) model for pancreas segmentation.
• Integrated two adversarial networks into a U-Net architecture to enhance prediction map accuracy.
• Incorporated attention blocks to preserve contextual information and improve segmentation performance.
• Implemented a backbone segmentor selection scheme and optimized attention block integration.

Main Results:

• The ADAU-Net achieved a 6.39% higher dice similarity coefficient compared to the baseline segmentation network.
• Demonstrated competitive performance against state-of-the-art methods on the NIH Pancreas-CT dataset.
• Experimental results validate the model's effectiveness in pancreas segmentation.

Conclusions:

• The ADAU-Net model provides accurate pancreas segmentation from CT images.
• The proposed method shows promise for clinical applications requiring precise organ delineation.