Forensic age estimation for pelvic X-ray images using deep learning.

Area of Science:

• Forensic anthropology
• Radiology
• Artificial intelligence in medicine

Background:

• Skeletal maturity assessment is crucial for forensic age estimation.
• Pelvic radiographs offer valuable indicators for bone age determination.
• Existing methods, like cubic regression, have limitations in automation and accuracy.

Purpose of the Study:

• To develop a deep learning model for bone age assessment using pelvic radiographs.
• To compare the performance of this deep learning model against the established cubic regression model for forensic age estimation.
• To evaluate the model's predictive ability for automated skeletal bone assessment.

Main Methods:

• A retrospective dataset of 1875 pelvic radiographs from individuals aged 10-25 years was utilized.
• A deep learning convolutional neural network model was developed for automated bone age estimation.
• Performance was evaluated by comparing mean absolute error (MAE) and root-mean-squared error (RMSE) against chronological age and the cubic regression model.

Main Results:

• The deep learning model achieved a mean MAE of 0.94 years and RMSE of 1.30 years for all test samples (10-25 years).
• For comparable age ranges (14-22 years), the deep learning model yielded MAE of 0.89 years and RMSE of 1.21 years.
• The existing cubic regression model showed a mean MAE of 1.05 years and RMSE of 1.61 years in the same age range.

Conclusions:

• The deep learning model demonstrates performance on par with the existing cubic regression model for bone age assessment.
• The developed model shows significant predictive ability for automated skeletal bone assessment using pelvic radiographs.
• Pelvic radiographs are valuable for bone age determination, and deep learning offers an effective automated solution.