Multi-scale U-like network with attention mechanism for automatic pancreas segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Deep neural networks have advanced automatic organ segmentation in abdominal CT scans.
• Automatic segmentation of small abdominal organs, like the pancreas, remains challenging due to anatomical variability and indistinct boundaries.

Purpose of the Study:

• To develop a more accurate automatic segmentation method for the pancreas in abdominal CT scans.
• To address the limitations of existing methods in segmenting small, anatomically variable organs.

Main Methods:

• A novel 2.5D U-net architecture incorporating an attention mechanism was proposed.
• The network utilizes both 2D and 3D convolutional layers to balance computational resources and spatial information capture.
• A cascaded framework was employed to further enhance segmentation accuracy.

Main Results:

• The proposed 2.5D U-net demonstrated superior performance in pancreas segmentation compared to state-of-the-art methods.
• Evaluation on the NIH pancreas dataset using the Dice similarity coefficient (DSC) confirmed the network's effectiveness.
• The attention mechanism and cascaded framework contributed to improved segmentation accuracy.

Conclusions:

• The developed 2.5D U-net with an attention mechanism offers a promising solution for accurate automatic pancreas segmentation.
• This approach effectively handles the challenges associated with segmenting small, variable abdominal organs.
• The method provides a valuable tool for medical imaging analysis and clinical applications.