Automated segmentation of vertebral cortex with 3D U-Net-based deep convolutional neural network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate segmentation of vertebral cortex is crucial for diagnosing spinal conditions.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Deep learning offers a potential solution for automated and efficient segmentation.

Purpose of the Study:

• To develop and evaluate a 3D U-Net deep learning model for automatic vertebral cortex segmentation.
• To compare the performance of the 3D U-Net model against other deep learning algorithms.

Main Methods:

• A 3D U-Net convolutional neural network was developed for automated vertebral cortical segmentation.
• Ten-fold cross-validation and data augmentation were employed on 1,672 chest CT scans.
• Segmentation performance was compared using Dice Similarity Coefficient (DSC), mean Intersection over Union (mIoU), Mean Percentage Accuracy (MPA), and Frames Per Second (FPS) against Res U-Net, Ki U-Net, and Seg Net.

Main Results:

• The 3D U-Net model achieved superior DSC (0.71 ± 0.03), mIoU (0.74 ± 0.08), and MPA (0.83 ± 0.02) compared to other methods.
• The model demonstrated promising automated segmentation accuracy for vertebral cortical bone.
• Frames Per Second (FPS) was slightly lower than Seg Net, indicating a trade-off between accuracy and speed.

Conclusions:

• The 3D U-Net deep learning model effectively segments vertebral cortical bone from CT images.
• This automated approach shows significant potential for improving efficiency and accuracy in radiological assessments.
• Further optimization may enhance processing speed while maintaining high segmentation performance.