Automated radiography assessment of ankle joint instability using deep learning

Area of Science:

• Radiology
• Artificial Intelligence
• Orthopedics

Background:

• Ankle joint instability diagnosis relies on measuring talar tilt and anterior talar translation from weight-bearing ankle radiographs.
• Manual measurement of these parameters can be subjective and time-consuming.
• Developing automated systems can improve objectivity and efficiency in diagnosing ankle instability.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) based system for automated measurement of talar tilt and anterior talar translation.
• To assess the accuracy and reliability of the DL system compared to clinician assessments.
• To determine the system's utility in supporting the clinical diagnosis of ankle instability.

Main Methods:

• A deep learning system was developed and trained using 1,452 anteroposterior and 2,984 lateral weight-bearing ankle radiographs from 4,000 patients.
• Exclusion criteria included patients with prior joint fusion, bone grafting, or joint replacement.
• Statistical analyses, including correlation coefficients and Bland-Altman plots, were used to compare DL-derived measurements with clinician assessments.

Main Results:

• The DL system demonstrated high accuracy for talar tilt measurements (e.g., Pearson's r = 0.798, ICC = 0.797).
• The system also showed high accuracy for anterior talar translation measurements (e.g., Pearson's r = 0.862, ICC = 0.861).
• The DL-calculated measurements showed strong agreement and consistency with clinician-assessed values.

Conclusions:

• The developed deep learning system provides objective and reproducible measurements of talar tilt and anterior talar translation.
• The system's high accuracy supports its potential to aid in the clinical interpretation of ankle instability.
• This AI tool can enhance routine radiographic practice for diagnosing ankle joint instability.