Automated bone age assessment from knee joint by integrating deep learning and MRI-based radiomics

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Traditional bone age assessment (BAA) methods carry risks of radiation exposure.
• There is a need for non-invasive and automated BAA techniques.
• Multiparametric knee MRI offers rich data for advanced analysis.

Purpose of the Study:

• To develop and evaluate a deep learning radiomics method for non-invasive and automated BAA using multiparametric knee MRI.
• To compare the performance of the deep learning model against traditional machine learning approaches.
• To assess the model's accuracy across different age groups and thresholds.

Main Methods:

• A retrospective study of 598 patients (age 10.00-29.99 years) using knee MRI (T1WI, T2*WI, PDWI).
• Three-dimensional convolutional neural networks (3D CNNs) were employed to extract and fuse radiomic features for age estimation.
• The deep learning model was compared with traditional machine learning models using hand-crafted features.

Main Results:

• The 3D CNN model achieved a mean absolute error of 1.32 ± 1.01 years for ages 10-25 years.
• Classification accuracies and AUCs exceeded 0.91 and 0.96, respectively, for key age thresholds.
• The model demonstrated higher accuracy than other MRI modalities and hand-crafted feature models, with a limitation noted for individuals over 25 years.

Conclusions:

• Deep learning radiomics enables non-invasive and automated bone age assessment from multimodal knee MRI.
• The 3D CNN approach shows potential for assisting radiologists and medicolegal experts in age estimation.
• Further refinement may be needed for accurate BAA in individuals over 25 years.