Skeletal bone age prediction based on a deep residual network with spatial transformer.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Manual bone age assessment from hand X-rays is time-consuming and prone to human error.
• Current methods involve significant workload and resource consumption.
• Radiologist proficiency greatly influences the accuracy and consistency of manual bone age estimation.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated bone age prediction.
• To improve the efficiency and accuracy of bone age assessment using X-ray imaging.
• To compare the performance of deep learning methods against traditional techniques.

Main Methods:

• Utilized deep neural networks for automated feature extraction from left-hand X-ray images.
• Employed convolutional neural networks (CNNs) for bone age assessment.
• Implemented and evaluated a model based on the ResNet architecture.

Main Results:

• Deep learning methods, particularly CNNs, outperform traditional image analysis techniques in feature extraction.
• The ResNet-based model achieved an average absolute error of 0.455 in bone age prediction.
• The developed model demonstrated a high prediction accuracy of up to 97.6%.

Conclusions:

• CNN-based feature extraction offers superior performance for bone age regression models compared to traditional machine learning.
• Automated assessment using deep learning enhances the accuracy of image-based bone age evaluation.
• The study highlights the potential of AI in improving diagnostic efficiency and reliability in radiology.