Automated Assessment of Bone Age Using Deep Learning and Gaussian Process Regression

Area of Science:

• Pediatric Endocrinology
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Bone age assessment is crucial for evaluating skeletal maturity in pediatric growth disorders.
• Current methods rely on physician interpretation of hand radiographs, leading to significant inter-observer and intra-observer variability.
• Existing reference models for bone age assessment have limitations in interpretation.

Purpose of the Study:

• To develop and validate a novel, automated method for bone age assessment.
• To enhance the accuracy and reduce variability in skeletal maturity estimation.
• To assist physicians in more precise bone age estimations.

Main Methods:

• A hybrid approach combining deep learning and Gaussian process regression was developed.
• The method leverages deep learning's sensitivity to image transformations (rotations, flips).
• Retrospective validation was performed on 12,611 hand radiographs from patients aged 0-19 years.

Main Results:

• The combined deep learning and Gaussian process regression model demonstrated improved predictive performance.
• Exploiting deep learning's sensitivity to image variations enhanced overall accuracy.
• The automated method shows potential for more consistent bone age assessments.

Conclusions:

• The novel automated bone age assessment method offers a promising solution to reduce variability in skeletal maturity estimation.
• Combining deep learning with Gaussian process regression enhances predictive accuracy.
• This approach can serve as a valuable tool to support clinical decision-making in pediatric growth disorders.