A Deep Automated Skeletal Bone Age Assessment Model with Heterogeneous Features Learning

Area of Science:

• Medical imaging analysis
• Pediatric endocrinology
• Machine learning applications

Background:

• Skeletal bone age assessment is crucial for pediatric disease detection and growth prediction.
• Manual assessment of hand and wrist X-rays is subjective, inefficient, and lacks consistent accuracy.
• Existing automated methods show insufficient accuracy for clinical application.

Purpose of the Study:

• To develop a highly accurate automated skeletal bone age assessment model.
• To overcome the limitations of conventional manual and current automated methods.
• To leverage deep learning for enhanced diagnostic capabilities in pediatric endocrinology.

Main Methods:

• Development of a deep learning framework combining Convolutional Neural Networks (CNNs) and Support Vector Regression (SVR).
• Utilized Multiple Kernel Learning (MKL) to process heterogeneous features, including hand/wrist X-ray images, race, and gender.
• Trained and validated the model on two distinct datasets.

Main Results:

• The proposed model demonstrated superior performance and higher bone age assessment accuracy compared to state-of-the-art methods.
• Fusion of heterogeneous features (radiographic and demographic) significantly improved assessment precision.
• The model achieved better description of bone maturation stages.

Conclusions:

• The deep automated model offers a more accurate and efficient approach to skeletal bone age assessment.
• Integrating diverse data sources enhances the predictive power for pediatric growth and disease.
• This framework holds significant potential for clinical application in pediatric endocrinology.