Ensemble Learning with Multiclassifiers on Pediatric Hand Radiograph Segmentation for Bone Age Assessment.

Area of Science:

• Medical Imaging
• Computer Vision
• Pediatric Radiology

Background:

• Accurate bone age assessment (BAA) in children is crucial for diagnosing growth disorders.
• Hand radiograph segmentation is a critical but challenging step in automated BAA.
• Existing segmentation methods lack the required precision and efficiency for clinical application.

Purpose of the Study:

• To develop a highly precise and efficient automatic segmentation method for pediatric hand radiographs.
• To improve the accuracy of bone age assessment (BAA) through enhanced radiograph segmentation.

Main Methods:

• Hand radiograph segmentation framed as a classification problem, predicting optimal thresholds.
• Utilized normalized histogram, mean, and variance as input features for ensemble learning classifiers.
• Trained and validated the model on 600 pediatric left-hand radiographs (age 1-18).

Main Results:

• The proposed method outperformed traditional techniques and U-Net in precision and computational load.
• Achieved high performance metrics: PSNR (52.43 dB), SSIM (0.97), DSC (0.97), and JSI (0.91).
• Demonstrated an average BAA performance improvement of at least 13% post-segmentation.

Conclusions:

• The developed automatic segmentation method is suitable for clinical application in pediatric BAA.
• High-precision hand radiograph segmentation significantly enhances bone age assessment accuracy.
• This approach offers a promising advancement for automated radiological diagnostics in pediatrics.