Region fine-grained attention network for accurate bone age assessment

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Endocrinology

Background:

• Bone age assessment is crucial for evaluating adolescent growth and development.
• Traditional methods using hand radiography face challenges due to bone variability, background masking, and high similarity between successive age radiographs.
• Accurate assessment is vital for timely medical intervention and treatment planning.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for precise bone age assessment from hand radiographs.
• To address the limitations of existing methods in distinguishing skeletal regions and handling visually similar images.
• To improve the accuracy and reliability of bone age estimation in pediatric populations.

Main Methods:

• A region fine-grained attention network (RFGA-Net) was proposed, incorporating a region aware attention (RAA) module and a fine-grained feature attention (FFA) module.
• The RAA module was designed to differentiate skeletal structures from the background by analyzing global spatial dependencies.
• The FFA module was developed to identify critical fine-grained feature regions, aiding in the differentiation of similar bone radiographs.

Main Results:

• The RFGA-Net demonstrated superior performance on the Radiological Society of North America (RSNA) pediatric bone dataset.
• The model achieved a mean absolute error (MAE) of 3.34.
• The model achieved a root mean square error (RMSE) of 4.02, indicating high accuracy in bone age estimation.

Conclusions:

• The proposed RFGA-Net effectively overcomes the challenges associated with bone age assessment from hand radiographs.
• The attention-based modules enable better distinction of skeletal regions and identification of subtle differences in bone development.
• RFGA-Net represents a significant advancement in automated bone age assessment, offering improved accuracy for clinical applications.