Automatic prosthetic-parameter estimation from anteroposterior pelvic radiographs after total hip arthroplasty using...

Area of Science:

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

Background:

• Post-operative X-ray imaging is crucial for total hip arthroplasty (THA) follow-ups.
• Manual parameter measurements from these radiographs are time-consuming and prone to variability.
• Automating this process can enhance efficiency and consistency in patient care.

Purpose of the Study:

• To introduce BKNet, a deep learning tool designed for automated landmark localization and parameter measurement in post-THA radiographs.
• To evaluate the performance of BKNet in comparison to human observers for accuracy, repeatability, and efficiency.

Main Methods:

• Utilized a dataset of 3072 radiographs from 3021 patients who underwent THA.
• Applied BKNet for automated landmark identification and parameter quantification on the selected radiographs.
• Assessed BKNet's performance using Bland-Altman analysis and compared its repeatability with that of human observers for 10 key parameters.

Main Results:

• BKNet achieved 75-percentile cut-off errors of less than 0.5 cm for all critical landmark points.
• Bland-Altman analysis demonstrated good agreement between BKNet's predicted parameters and the ground truth.
• The deep learning model matched the repeatability of human observers for 7 out of 10 measured prosthetic parameters.

Conclusions:

• BKNet demonstrates accuracy comparable to human observers in estimating prosthetic parameters from post-THA X-rays.
• The deep learning approach offers significant advantages in cost-effectiveness, repeatability, and time savings compared to manual measurements.
• BKNet represents a valuable tool for efficient and reliable post-operative assessment in total hip arthroplasty.