ACEnet: Anatomical context-encoding network for neuroanatomy segmentation.

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Brain structure segmentation from MR scans is crucial for quantifying brain morphology.
• 3D deep learning models offer accuracy but are computationally expensive.
• Existing 2D deep learning methods lack sufficient 3D spatial context for precise segmentation.

Purpose of the Study:

• To develop an efficient and accurate 2D deep learning method for brain structure segmentation.
• To incorporate 3D spatial and anatomical context into 2D convolutional neural networks (CNNs).
• To improve the computational efficiency and segmentation accuracy compared to existing methods.

Main Methods:

• Developed an Anatomical Context-Encoding Network (ACEnet) utilizing 2D CNNs.
• Incorporated an anatomical context encoding module to integrate anatomical information.
• Integrated a spatial context encoding module to incorporate 3D image information.
• Utilized a skull stripping module to focus the network on brain regions.

Main Results:

• ACEnet demonstrated promising performance on three benchmark datasets.
• The method achieved competitive segmentation accuracy for brain structures.
• ACEnet offered significant computational efficiency compared to 3D methods.

Conclusions:

• ACEnet effectively integrates 3D spatial and anatomical contexts into 2D CNNs for brain segmentation.
• The proposed network provides an efficient and accurate solution for brain morphology quantification.
• ACEnet represents a significant advancement in automated brain structure analysis from MR imaging.