Development and clinical validation of a deep learning-based knee CT image segmentation method for robotic-assisted...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in surgery
• Orthopedic surgery

Background:

• Developing precise and rapid knee joint CT image segmentation is crucial for robotic-assisted total knee arthroplasty (TKA).
• Existing methods may lack the required accuracy and speed for clinical application.

Purpose of the Study:

• To develop and validate the Dual-path Double Attention Transformer (DDA-Transformer), a novel deep convolutional neural network for knee joint CT image segmentation.
• To assess the performance and clinical utility of DDA-Transformer in robotic-assisted TKA.

Main Methods:

• The DDA-Transformer model was developed for segmenting femoral, tibial, patellar, and fibular components in knee CT images.
• Segmentation performance (accuracy, speed) and clinical validation in robotic-assisted TKA were evaluated.
• Comparison with six other deep learning networks including nnUnet, TransUnet, and 3D-Unet.

Main Results:

• DDA-Transformer demonstrated superior performance over six other networks, evidenced by higher Dice coefficients, intersection over union, and lower average surface distance and Hausdorff distance.
• Significantly faster segmentation speeds were observed for DDA-Transformer compared to nnUnet, TransUnet, and 3D-Unet (p < 0.01).
• Robotic-assisted TKA utilizing DDA-Transformer showed improved surgical accuracy compared to the manual group.

Conclusions:

• The DDA-Transformer network offers significantly enhanced accuracy and robustness for knee joint CT image segmentation.
• This deep learning approach provides a convenient and stable solution for knee joint segmentation, leading to improved accuracy in TKA procedures.