Comparing two different automatic methods to measure femoral neck-shaft angle based on PointNet++ network

Area of Science:

• Orthopedics
• Medical Imaging
• Computer-Aided Surgery

Background:

• Accurate femoral neck-shaft angle (NSA) measurement is vital for diagnosing hip conditions and planning total hip arthroplasty.
• Manual NSA measurement lacks repeatability, and discrepancies between 2D and 3D imaging methods are not well-defined.
• Computer-aided methods offer a solution for automated and precise NSA measurement.

Purpose of the Study:

• To develop and validate accurate, automatic computer-aided methods for measuring the femoral neck-shaft angle (NSA).
• To compare the consistency and accuracy of 2D projection and 3D model-based NSA measurements.
• To evaluate the performance of deep learning-based segmentation for femoral component analysis.

Main Methods:

• Utilized PointNet++ to segment femoral point cloud data from 310 subjects into head, neck, and shaft regions.
• Developed a 2D measurement method by simulating anteroposterior radiographs and fitting axes on a projection plane.
• Developed a 3D measurement method by directly fitting the femoral neck and shaft axes in spatial coordinates.

Main Results:

• Achieved high accuracy in femoral segmentation with a Dice coefficient of 0.9746 and MIoU of 0.9165.
• No significant difference was observed between the automatic 2D, automatic 3D, and manual NSA measurement methods.
• The 2D and 3D methods demonstrated an average accuracy of 98.00% and an average error of 2.58°.

Conclusions:

• Proposed two accurate and automatic methods for NSA measurement using 2D and 3D approaches.
• The developed computer-aided methods provide reliable NSA measurements for clinical applications.
• Results indicate that 2D and 3D NSA measurements are consistent, despite non-intersecting axes in 3D space.