Reclassification of Demirjian's mandibular premolars staging for age estimation based on semi-automated segmentation of deep convolutional neural network
The accuracy of age estimation using radiographic approach is important in forensic application. Despite its inconsistency, Demirjian method has been widely used to estimate age in children and young adults. This study aims to develop a semi-automated segmentation of dental development based on fi...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Elsevier Ltd.
2021
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/46019/1/Reclassification%20of%20Demirjian%E2%80%99s%20-%20Copy.pdf http://ir.unimas.my/id/eprint/46019/ https://www.sciencedirect.com/science/article/abs/pii/S2666225621000117 https://doi.org/10.1016/j.fri.2021.200440 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The accuracy of age estimation using radiographic approach is important in forensic application. Despite its
inconsistency, Demirjian method has been widely used to estimate age in children and young adults. This study
aims to develop a semi-automated segmentation of dental development based on first (P1) and second (P2)
permanent mandibular premolars and to correlate the mean age of the novel staging method with Demirjian. A
stratified sample of six-hundred-fifty-seven panoramic radiographs was retrospectively collected. All training
datasets underwent image segmentation process to form training dataset for the classifier. By implementing the
Deep Convolution Neural Network (DCNN), 80% of the dataset was assigned for training and validation while the
remaining 20% for testing. Classification performance was measured using confusion matrix. With the maximum
number of 10-epochs assigned in DCNN, 92.5% of classification accuracy was obtained. On the tested image,
differences between predicted age (PA) of the digital processing and the chronological age (CA) were calculated
and expressed in mean error. The PA for P1 is underestimated by 0.17 (95% CI: 0.16-0.51) in male while female
is overestimated by 0.22 (95% CI: 0.51-0.07). P2 exhibited better accuracy with an overestimation of 0.02 (95%
CI: 0.24-0.20) and 0.03 (95% CI: 0.34-0.27) for male and female, respectively. A new sub-stage of stage D was
introduced. The overall performance of the presented classification method to stage premolars development on
panoramic radiographs was superior as compared to the original method, and modification of mandibular
premolar staging in Demirjian method is established in this study. |
|---|