A ROI Extraction Method for Wrist Imaging Applied in Smart Bone-Age Assessment System

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Accurate Bone Age Assessment (BAA) is crucial for monitoring teenage growth and development.
• Current BAA methods struggle with precise reference bone extraction from carpal bones due to irregular shapes and proportions.
• Inaccurate reference bone identification leads to lower BAA accuracy.

Purpose of the Study:

• To propose an optimized YOLO model for accurate Region of Interest (ROI) extraction in wrist X-ray images for BAA.
• To improve the detection accuracy of reference bones, especially those with irregular shapes.
• To enhance the overall accuracy and efficiency of Bone Age Assessment.

Main Methods:

• Developed YOLO-DCFE, an optimized YOLO model integrating Deformable convolution-focus (Dc-focus), Coordinate attention (Ca) module, Feature level expansion, and Efficient Intersection over Union (EIoU) loss.
• Utilized a dataset of 10041 wrist X-ray images captured by professional medical cameras.
• Evaluated the model's performance in terms of detection speed and accuracy.

Main Results:

• YOLO-DCFE achieved a 99.8% detection accuracy for all ROIs, surpassing other models.
• The model demonstrated superior detection speed, reaching 16 Frames Per Second (FPS).
• The proposed method effectively extracts features of irregular reference bones and reduces misdiscrimination.

Conclusions:

• YOLO-DCFE significantly improves the accuracy and efficiency of reference bone extraction for Bone Age Assessment.
• The optimized YOLO model offers a promising solution for enhancing BAA in clinical practice.
• This advancement contributes to more reliable monitoring of adolescent growth and development.