Deep Learning Framework for Automatic Bone Age Assessment.

Area of Science:

• Pediatric Radiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Bone age assessment is crucial for identifying endocrine and metabolic disorders in children.
• Skeletal age provides a more accurate measure of biological growth than chronological age.
• Current methods often rely on manual interpretation of hand X-rays, which can be subjective.

Purpose of the Study:

• To develop a fully automated deep learning approach for bone age assessment.
• To improve the accuracy and consistency of bone age predictions.
• To provide an AI-based supplementary tool for clinical decision-making.

Main Methods:

• Utilized a dataset from the 2017 Pediatric Bone Age Challenge.
• Employed transfer learning with the pre-trained InceptionV3 neural network architecture.
• Trained the model on left hand X-ray images tagged with patient age and gender.

Main Results:

• The automated deep learning model achieved a mean absolute error of 5.921 months in bone age prediction.
• Demonstrated high agreement among automated methods for evaluating X-rays.
• The InceptionV3 architecture proved effective for this task.

Conclusions:

• A fully automated deep learning system can accurately predict bone age from hand X-rays.
• This AI-driven approach can serve as an effective assistive tool for clinicians.
• The method shows promise for enhancing the diagnosis of developmental abnormalities.