Automatic estimation of hallux valgus angle using deep neural network with axis-based annotation.

Area of Science:

• Orthopedic imaging analysis
• Artificial intelligence in medicine
• Radiographic measurement techniques

Background:

• Hallux valgus angle (HVA) and intermetatarsal angle (IMA) are critical metrics in diagnosing foot deformities.
• Manual measurement of HVA and IMA on radiographs can be subjective and time-consuming.
• Developing automated methods for accurate radiographic analysis is essential for efficient clinical practice.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) model for automatic measurement of HVA and IMA.
• To assess the accuracy of the DNN model by comparing its measurements to those of experienced foot and ankle surgeons.
• To evaluate the model's performance against established inter-rater reliability standards.

Main Methods:

• A DNN model was trained to identify bone axes on foot radiographs for HVA and IMA calculation.
• The model was developed using 1798 radiographs from a combined cohort.
• Retrospective validation was performed on 92 radiographs, comparing DNN measurements to the median of three surgeons' manual measurements.

Main Results:

• The DNN model achieved a mean absolute error (MAE) of 1.3° for HVA, significantly lower than the surgeons' inter-rater difference (2.0°).
• For IMA, the model's MAE was 0.8°, showing no significant difference compared to the surgeons' inter-rater difference (1.0°).
• The model demonstrated high accuracy and reliability in measuring both HVA and IMA.

Conclusions:

• The developed DNN model accurately measures HVA and IMA on foot radiographs.
• The model's accuracy is comparable to that of specialist foot and ankle surgeons.
• This automated approach offers a reliable alternative for radiographic assessment of foot alignment.