[Paper Review] Forensic Dental Age Estimation Using Modified Deep Learning Neural Network

İsa Ataş, Cüneyt Özdemir+2 more|arXiv (Cornell University)|Aug 21, 2022

Dental Radiography and Imaging35 references4 citations

TL;DR

This study proposes a modified InceptionV3 deep learning model for automated forensic dental age estimation using panoramic radiographs (DPR) from individuals aged 8–68. By leveraging transfer learning and architectural pruning, the method achieves a mean absolute error (MAE) of 3.13 years, root mean square error (RMSE) of 4.77, and R² of 87.2%, demonstrating high accuracy and efficiency in age prediction without requiring manual feature extraction.

ABSTRACT

Dental age is one of the most reliable methods to identify an individual's age. By using dental panoramic radiography (DPR) images, physicians and pathologists in forensic sciences try to establish the chronological age of individuals with no valid legal records or registered patients. The current methods in practice demand intensive labor, time, and qualified experts. The development of deep learning algorithms in the field of medical image processing has improved the sensitivity of predicting truth values while reducing the processing speed of imaging time. This study proposed an automated approach to estimate the forensic ages of individuals ranging in age from 8 to 68 using 1,332 DPR images. Initially, experimental analyses were performed with the transfer learning-based models, including InceptionV3, DenseNet201, EfficientNetB4, MobileNetV2, VGG16, and ResNet50V2; and accordingly, the best-performing model, InceptionV3, was modified, and a new neural network model was developed. Reducing the number of the parameters already available in the developed model architecture resulted in a faster and more accurate dental age estimation. The performance metrics of the results attained were as follows: mean absolute error (MAE) was 3.13, root mean square error (RMSE) was 4.77, and correlation coefficient R$^2$ was 87%. It is conceivable to propose the new model as potentially dependable and practical ancillary equipment in forensic sciences and dental medicine.

Motivation & Objective

• To develop an automated, accurate, and efficient method for forensic dental age estimation using dental panoramic radiographs (DPR) to reduce reliance on time-consuming manual techniques.
• To address the limitations of traditional age estimation methods, including subjectivity, measurement bias, and high labor intensity.
• To improve age prediction accuracy across a broad age range (8–68 years), particularly in post-pubertal individuals where tooth development slows.
• To optimize transfer learning by modifying the InceptionV3 architecture to reduce parameters and enhance speed and performance.
• To validate the model’s focus areas using Grad-CAM heatmaps, ensuring interpretability and alignment with anatomical regions critical to age estimation.

Proposed method

• Employed transfer learning using pre-trained InceptionV3, DenseNet201, EfficientNetB4, MobileNetV2, VGG16, and ResNet50V2 models on 1,332 DPR images.
• Selected InceptionV3 as the best-performing base model and modified its architecture to reduce parameters, resulting in the InceptionV3Mixed_04 variant.
• Applied data augmentation techniques during preprocessing to improve model generalization and robustness.
• Performed feature extraction using the modified InceptionV3Mixed_04 model on DPR images without manual segmentation or feature engineering.
• Used linear regression to map extracted features to chronological age, enabling end-to-end regression-based age estimation.
• Generated Grad-CAM heatmaps to visualize and interpret the model’s attention to specific anatomical regions, such as the maxilla and upper teeth.

Experimental results

Research questions

• RQ1 Can a modified transfer learning-based deep neural network achieve high accuracy in dental age estimation across a broad age range (8–68 years) using DPR images?
• RQ2 How does architectural pruning of a pre-trained InceptionV3 model affect performance, inference speed, and parameter efficiency in dental age estimation?
• RQ3 What anatomical regions in DPR images does the deep learning model primarily focus on during age prediction, and are these regions consistent with clinical knowledge of age-related dental changes?
• RQ4 How does the performance of the proposed modified InceptionV3 model compare to existing methods in terms of MAE, RMSE, and R² across diverse age groups?
• RQ5 To what extent does the model maintain accuracy in older individuals (over 55 years), and what factors may contribute to performance degradation in this group?

Key findings

• The modified InceptionV3Mixed_04 model achieved a mean absolute error (MAE) of 3.13 years, indicating high precision in age estimation across the 8–68 age range.
• The root mean square error (RMSE) was 4.77, reflecting the model’s overall prediction error distribution and robustness.
• The coefficient of determination (R²) reached 87.2%, demonstrating strong correlation between predicted and actual chronological ages.
• The model outperformed existing methods in terms of accuracy across a wide age range, particularly in the 16–65 age group where prior studies reported MAE values of 6.48–8.58.
• Grad-CAM visualization confirmed that the model primarily focused on the maxilla and upper teeth, aligning with clinically relevant regions for age estimation.
• Performance degradation in individuals over 55 years was observed, likely due to reduced dental development changes and limited training data in that age group.