Automatic Age Estimation and Majority Age Classification From Multi-Factorial MRI Data.

Area of Science:

• Radiology
• Forensic Science
• Artificial Intelligence

Background:

• Accurate age estimation is crucial in clinical and forensic contexts.
• Current methods often have limitations in age range, accuracy, and objectivity.
• Existing multi-factorial approaches may rely on ionizing radiation and subjective assessments.

Purpose of the Study:

• To develop an automatic, multi-factorial age estimation method using MRI data.
• To extend the age assessment range beyond the typical 19 years using hand, clavicle, and teeth data.
• To improve upon existing methods by reducing subjectivity and radiation dependence.

Main Methods:

• Utilized a deep convolutional neural network (CNN) for age estimation.
• Integrated MRI data from hand, clavicle, and wisdom teeth.
• Trained the CNN on a dataset of 322 subjects aged 13-25 years.

Main Results:

• Achieved a mean absolute prediction error of 1.01±0.74 years for chronological age regression.
• Demonstrated superior performance of a thresholded regression classifier over a direct classification loss approach for majority age classification.
• Showcased the method's effectiveness in minimizing misclassification of minors as adults.

Conclusions:

• The proposed multi-factorial MRI method effectively extends age estimation to 25 years.
• This approach overcomes limitations of traditional methods, including radiation use and subjectivity.
• The integrated deep learning model provides a robust and objective tool for age assessment.