A Deep Learning Tool for Hip Minimum Joint Space Width Calculation on Antero-posterior Pelvis Radiographs

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Minimum joint space width (mJSW) is a key quantitative metric for hip osteoarthritis progression.
• Continuous mJSW measurement offers advantages over traditional categorical classifications.

Purpose of the Study:

• To develop an automated algorithm for measuring mJSW in native hips.
• To enable accurate mJSW assessment on antero-posterior (AP) pelvis radiographs.

Main Methods:

• An end-to-end algorithm combining deep learning segmentation and computer vision was created.
• A dataset of 300 radiographs was annotated for model training and validation.
• External validation was performed on 75 images from the Osteoarthritis Initiative (OAI).

Main Results:

• The algorithm achieved a mean absolute error of 0.87 ± 1.05 mm compared to human measurements.
• 70% of algorithm measurements were within 1 mm of human measurements.
• External validation on OAI data showed a mean absolute error of 0.86 ± 0.69 mm.

Conclusions:

• An automated model for mJSW measurement on AP pelvis radiographs has been successfully developed and validated.
• The algorithm demonstrates sub-millimeter accuracy, potentially improving longitudinal patient monitoring.
• This tool can streamline clinical research on the natural history of hip joint osteoarthritis.