All evidence
Clinical Validation of an Artificial Intelligence Software for Bone Age Assessment based on Greulich & Pyle method in a Portuguese Paediatric Cohort.
Data
267 hand and wrist radiographs from 229 children
79 radiographs classified as pathological
Mean chronological age = 9.16 years
Design
Standalone performance of BoneAge
Ground truth established by the consensus of two expert readers based on the G&P atlas
In case of discrepancies the radiographs were reviewed by an experienced radiologist with 10 years of experience
Results
G&P prediction
BoneAge
Bias -2.55 mo
MAE 4.9 mo
r 0.98
Classification into healthy or pathological
BoneAge
Cohen’s d 0.667
Sensitivity 67.6%
Specificity 96.2%
PPV 92.4%
NPV 81.4%
Accuracy 84.6%
