Hybrid approach with convolutional neural network and regression models for age estimation in panoramic radiographs

Determining chronological age is a relevant aspect of criminal and civil interest for identifying corpses and refugees. This study aimed to estimate chronological age using a hybrid model that combines a convolutional neural network (CNN) with regression models based on panoramic radiographs.

This observational, retrospective and cross-sectional study analyzed 2,546 panoramic radiographs from a Brazilian practice. Panoramic radiographs of patients over 18 years old with healthy, erupted lower premolars were selected. The radiographs were segmented using GIMP, and the dataset was divided into three sets: 70% for model training, 15% for validation, and 15% for testing. Based on the MobileNetV2 architecture, the CNN model extracted 576 attributes from each image, which were then used in conjunction with a decision-tree regression model to estimate age. To evaluate the model's performance, metrics such as Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Coefficient of Determination and Mean Square Error were used.

3,423 teeth were analyzed, and the patients' mean age was 32.65 years. The overall MAE was 6.52 years (R² = 0.67); however, for the age group 31 to 35, the hybrid AI model had an MAE of 0.29 years for the male sample and 0.16 years for the female sample.

The hybrid AI model proved accurate and robust for patients aged 31-35 years; however, in other age groups, the model performed unsatisfactorily, particularly in the 40-50 and 51-55 age ranges.