Proximal femur parameter measurement via improved PointNet++

Abstract

Background Femoral morphological studies and parameter measurements play a crucial role in diagnosing hip joint disease, preoperative planning for total hip arthroplasty, and prosthesis design. Doctors usually perform parameter measurements manually in clinical practice, but it is time‐consuming and labor‐intensive. Moreover, the results rely heavily on the doctor's experience, and the repeatability is poor. Therefore, the accurate and automatic measurement methods of proximal femoral parameters are of great value. Method We collected 300 cases of clinical CT data of the femur. We introduced the adaptive function adjustment module to the neural network PointNet++ to strengthen the global feature extraction of the point cloud for improving the accuracy of femur segmentation. We used the improved PointNet++ network to segment the femur into three parts: femoral head, femoral neck, and femoral shaft. We evaluated the segmentation accracy using Dice Coefficient, MIoU, recall, and precision indicators. We achieved the automatic measurement of the proximal femoral parameters using the shape fitting algorithms, and compared the automatic and manual measurement results. Results The Dice, MIoU, recall and precision indicator of the improved segmentation algorithm reached 98.05%, 96.55%, 96.63%, and 96.03%, respectively. The comparison between automatic and manual measurement results showed that the mean accuracies of all parameters were above 95%, the mean errors were less than 5 mm and 3°, and the ICC values were more than 0.8, indicating that the automatic measurement results were accurate. Conclusion Our improved PointNet++ network provided high‐precision segmentation of the femur. We further completed automatic measurement of the femur parameters and verified its high accuracy. This method is of great value for the diagnosis and preoperative planning of hip diseases.

Tree Species Classification Based on PointNet++ and Airborne Laser Survey Point Cloud Data Enhancement

“…For example, Xin [35] used X-ray computed tomography to obtain 3D volume data of metal powder particles, classified them into six categories using PointNet++, and achieved an accuracy of 93.8%. Further, Yang [36] collected 300 clinical CT data points on femurs and used the improved PointNet++ network to divide femurs into three parts: femoral head, neck, and shaft, and acquired a result accuracy of >95%. Jing [37] integrated the Squeeze-and-Excitation (SE) attention mechanism into PointNet++ for multispectral LiDAR point cloud classification tasks and used the PointNet++ model to classify roads, buildings, grasslands, trees, soils, and power lines, achieving an overall accuracy of 91.16%.…”