Automating Angle Measurements on Foot Radiographs in Young Children: Feasibility and Performance of a Convolutional...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate measurement of foot angles on radiographs is crucial for evaluating malalignment in pediatric patients.
• Traditional manual measurement methods are time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a Convolutional Neural Network (CNN) model for automated angle measurement on foot radiographs.
• To compare the accuracy and efficiency of the CNN model against manual measurements by experienced radiologists.

Main Methods:

• A retrospective study utilized 450 foot radiographs from 216 children under 3 years old.
• A U-Net CNN model with a ResNet-34 backbone was employed for image segmentation.
• Automated angle calculations (talocalcaneal, talo-1st metatarsal) were performed using Simon's approach and compared to manual measurements by two pediatric radiologists.

Main Results:

• The CNN model demonstrated high spatial overlap with manual segmentations (Dice coefficients 0.81-0.94).
• Angle measurement agreement between the CNN and radiologists was moderate to substantial (ICC 0.71-0.73 for lateral, 0.41-0.52 for AP views).
• Automated measurements were significantly faster (3±2s) than manual methods (114±24s), representing a 39-fold increase in speed.

Conclusions:

• A CNN model can effectively segment immature ossification centers and automatically calculate pediatric foot angles.
• The automated system provides comparable accuracy to manual methods while drastically improving efficiency.
• This AI-driven approach holds promise for streamlining radiographic analysis in pediatric orthopedics.