A deep learning-based computer-aided diagnosis method of X-ray images for bone age assessment

Area of Science:

• Radiology
• Artificial Intelligence
• Pediatric Endocrinology

Background:

• Bone age assessment from hand-wrist X-rays is crucial for diagnosing pediatric growth disorders.
• Current clinical assessments are subjective and depend on physician experience, impacting accuracy.
• Automating bone age assessment can enhance diagnostic consistency and patient-specific treatment planning.

Purpose of the Study:

• To develop a deep learning-based computer-aided diagnosis (CAD) method for objective bone age assessment.
• To reduce the need for expensive manual annotations through unsupervised learning for region localization.
• To improve the accuracy and reliability of bone age estimation in pediatric patients.

Main Methods:

• An unsupervised learning approach was used for informative region localization in hand-wrist X-rays.
• A pre-trained deep learning model (MobileNetV3) served as the backbone for feature extraction.
• A Multiple Layer Perceptron with gender information was employed for the final bone age prediction.
• An image-processing pipeline was developed to support the CAD system.

Main Results:

• The proposed deep learning method achieved a mean absolute error (MAE) of 6.2 months on the public RSNA dataset.
• An MAE of 5.1 months was obtained on an additional dataset, demonstrating robust performance.
• The integration of gender information improved prediction accuracy, aligning with clinical observations.

Conclusions:

• The developed deep learning CAD system offers a reliable and accurate method for bone age assessment.
• This approach can assist clinicians in diagnosing growth disorders and personalizing treatment strategies.
• Unsupervised learning for region localization effectively reduces annotation costs and enhances the CAD system's practicality.