Bone age assessment from articular surface and epiphysis using deep neural networks

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Endocrinology

Background:

• Bone age assessment is crucial for diagnosing genetic and endocrine disorders.
• Traditional methods rely on subjective radiologist interpretation of hand radiographs, leading to errors and delays.
• Existing AI methods lack focus on specific anatomical regions for improved bone age prediction.

Purpose of the Study:

• To develop a deep learning model for precise bone age prediction using articular surface and epiphysis regions.
• To enhance the accuracy and efficiency of bone age diagnosis in clinical settings.
• To address limitations of traditional and general AI-based bone age assessment techniques.

Main Methods:

• Established articular surface and epiphysis datasets from the RSNA pediatric bone age challenge.
• Manually segmented specific feature regions (articular surface, epiphysis) from hand radiographs.
• Employed five deep convolutional neural networks (ResNet50, SENet, DenseNet-121, EfficientNet-b4, CSPNet) for prediction.

Main Results:

• The best-performing model achieved a Mean Absolute Error (MAE) of 7.34 months on the specialized datasets.
• The deep learning approach demonstrated superior accuracy and speed compared to radiologists.
• The developed models offer a more efficient and reliable tool for bone age diagnosis.

Conclusions:

• Deep convolutional neural networks focused on articular surface and epiphysis provide accurate bone age prediction.
• This AI-driven method surpasses traditional radiologist assessments in accuracy and efficiency.
• The project offers a valuable tool for genetic and endocrine disease diagnosis in pediatrics.