A real-time automated bone age assessment system based on the RUS-CHN method

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Pediatric Endocrinology

Background:

• Bone age assessment (BAA) is crucial for evaluating pediatric physical growth and skeletal development.
• Traditional BAA methods involve time-consuming manual segmentation of hand radiographs or direct regression.
• Existing BAA systems often require significant computational resources and time for analysis.

Purpose of the Study:

• To develop an automated, end-to-end system for rapid and accurate bone age assessment.
• To improve the efficiency and reduce the computational demands of pediatric BAA.
• To provide physicians with reliable bone age estimations without the need for manual hand segmentation.

Main Methods:

• Utilized three real-time target detection models for identifying key bone grades and locations.
• Employed Key Bone Search (KBS) post-processing with the RUS-CHN approach for precise localization.
• Implemented a Lightgbm regression model for bone age prediction and evaluated performance using IOU, MAE, RMSE, and RMSPE.

Main Results:

• Achieved an average Intersection over Union (IOU) of at least 0.9 for key bone localization.
• The system demonstrated high accuracy with a Mean Absolute Error (MAE) of 0.35 years.
• Inference times were significantly reduced, with critical bone localization at 26 ms and bone age prediction at 2 ms on a GPU.

Conclusions:

• Developed a robust, automated end-to-end BAA system integrating real-time detection and regression.
• The system efficiently processes the RUS-CHN method, providing bone age and key bone information in real-time.
• This automated BAA system offers a stable, accurate, and fast alternative to traditional methods, aiding clinical decision-making.