A novel dual-branch segmentation algorithm for overall spine segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurosurgery

Background:

• Accurate identification of bone structures is crucial for surgical planning and minimizing damage.
• Automating the labeling of vertebrae in computed tomography (CT) images is a challenging but important task.
• Deep learning offers a promising avenue for automating complex medical image analysis tasks.

Purpose of the Study:

• To develop an automated method for labeling vertebrae in CT images.
• To improve the efficiency and accuracy of spinal structure identification for surgical guidance.
• To introduce the DBU-Net deep learning segmentation network within the nnUnet framework.

Main Methods:

• The DBU-Net incorporates a multi-scale feature channel attention module to integrate information from different image scales.
• A dual-branch decoder architecture, enhanced with a contextual Transformer module, captures global contextual information.
• Features from both branches interact at each decoding stage, merging global context with local details for improved segmentation.

Main Results:

• DBU-Net was evaluated on the Vertebrae Segmentation (VerSe) dataset (MICCAI 2019 and 2020).
• The network achieved state-of-the-art performance, with an average Dice coefficient of 94.59%.
• These results demonstrate the effectiveness of DBU-Net in segmenting spinal structures.

Conclusions:

• DBU-Net shows significant potential to aid surgeons in accurately identifying spinal structures.
• The automated segmentation can lead to more precise surgical execution and better disease diagnosis.
• This deep learning approach offers robust technical support for spine-related interventions.