CDUNeXt: efficient ossification segmentation with large kernel and dual cross gate attention.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurosurgery

Background:

• Ossification of the ligamentum flavum (OLF) is a primary cause of spinal stenosis.
• Current methods for identifying OLF rely on subjective clinical experience, leading to inefficiency and errors.
• Accurate and efficient identification of OLF is crucial for diagnosing spinal stenosis.

Purpose of the Study:

• To introduce a deep learning method for automated and efficient identification of ossified regions in the ligamentum flavum.
• To develop a lightweight and accurate segmentation network for OLF diagnosis.
• To address the clinical need for objective and reliable OLF assessment.

Main Methods:

• Proposed CDUNeXt, a lightweight deep learning model for OLF segmentation.
• Utilized large-kernel convolutions for capturing long-range feature dependencies.
• Incorporated dual-cross-gate-attention (DCGA) for sequential channel and spatial dependency capture.
• Focused on achieving fast and accurate segmentation with reduced parameters and complexity.

Main Results:

• CDUNeXt demonstrated superior segmentation performance compared to existing methods.
• The model achieved an optimal balance between lightweight design and computational cost.
• Achieved high accuracy in identifying ossified regions of the ligamentum flavum.

Conclusions:

• CDUNeXt offers a novel, efficient, and accurate deep learning solution for OLF diagnosis.
• This work pioneers the application of deep learning in OLF assessment, improving upon subjective methods.
• The developed lightweight network contributes to advancements in medical image segmentation and spinal stenosis research.