NCT05901857

Assessing the Precision of Convolutional Neural Networks for Dental Age Estimation From Panoramic Radiographs

Assessing the Precision of Convolutional Neural Networks for Dental Age Estimation in an Egyptian Population From Digital Panoramic Radiographs: A Diagnostic Accuracy Study

• Status: Unknown
• Phase: —
• Study type: Observational
• Enrollment: 22 (estimated)
• Sponsor: Cairo University · Academic / Other
• Sex: All
• Age: 6 Years – 16 Years
• Healthy volunteers: —

Summary

The aim of this study is to assess the accuracy of a convolutional neural network in dental age estimation from digital panoramic radiographs. The reference standard will be the chronological age of the patient.

Detailed description

Willems method is a dental age estimation technique modified from Demirjian method by creating new tables from which a maturity score is directly expressed in years. Panoramic radiographs of all participants will be taken with their informed consent, then they will be numbered and coded. Chronological age for each participant will be calculated by subtracting date of birth from date of radiograph and the real age will be blinded from the researcher (The chronological age is the ground truth). All panoramic radiographs will be examined twice by the main author to determine the dental age according to Willems method. The seven mandibular left teeth excluding the third molar will be scored as '0' for absence of calcification, and 'A' to 'H', depending on the stage of calcification. Each letter corresponds to a score which is the dental age fraction using tables for boys and girls. Summing the scores for the seven left mandibular teeth directly will result in the estimated dental age. The dental radiologist estimation accurancy will be compared to the ground truth (first index test). The second index test which will also be compared to the ground truth is the CNN model. To prepare the dataset for the CNN model, a rigorous preprocessing procedure will be followed. This will involve resizing the images to the desired dimensions, segmenting the teeth parts to be included in the image, and applying data augmentation techniques to enhance the quality and quantity of the dataset. The dataset will then be split into training and testing sets using a 20:80 ratio, which will be carefully selected based on the expected number of samples. Also the accuracy of the model will be assessed compared to the ground truth (the chronological ages).

Conditions

• Age Problem

Interventions

Type	Name	Description
DIAGNOSTIC_TEST	convolutional neural network	A deep learning model for dental age classification from panoramic images

Timeline

• Start date: 2023-06-30
• Primary completion: 2024-01-01
• Completion: 2025-12-01
• First posted: 2023-06-13
• Last updated: 2023-06-13

Locations

1 site across 1 country: Egypt