Proximal femur parameter measurement via improved PointNet+

Area of Science:

• Medical Imaging
• Computer Vision
• Orthopedic Surgery

Background:

• Femoral morphological studies are vital for diagnosing hip joint disease, planning total hip arthroplasty, and prosthesis design.
• Manual measurement of proximal femoral parameters is time-consuming, labor-intensive, and lacks repeatability.
• Accurate and automatic measurement methods for proximal femoral parameters are highly valuable.

Purpose of the Study:

• To develop an accurate and automatic method for measuring proximal femoral parameters.
• To improve the accuracy and efficiency of femoral segmentation for clinical applications.

Main Methods:

• Utilized 300 clinical CT datasets of the femur.
• Introduced an adaptive function adjustment module to PointNet++ for enhanced global feature extraction and femur segmentation.
• Segmented the femur into head, neck, and shaft using the improved PointNet++ network.
• Evaluated segmentation accuracy using Dice Coefficient, MIoU, recall, and precision.
• Performed automatic parameter measurement via shape fitting algorithms and compared with manual measurements.

Main Results:

• The improved segmentation algorithm achieved high accuracy with Dice (98.05%), MIoU (96.55%), recall (96.63%), and precision (96.03%).
• Automatic measurements demonstrated mean accuracies above 95%, with mean errors <5mm and <3°.
• Intraclass Correlation (ICC) values exceeded 0.8, confirming the accuracy of automatic measurements.

Conclusions:

• The improved PointNet++ network enables high-precision femur segmentation.
• Automatic measurement of femur parameters was successfully achieved with high accuracy.
• This automated approach offers significant value for diagnosing hip diseases and preoperative planning.