Deep learning for age estimation from panoramic radiographs: A systematic review and meta-analysis

Panoramic radiographs are widely used for age estimation in clinical and forensic domains. Conventionally, age estimation uses humans assessing tooth development and deducing the expected age from that. Deep learning may improve or substitute this traditional approach and allow age estimation at scale in routine settings. The objective of this systematic review was to assess the performance of deep learning for age estimation on panoramic radiographs.

Studies using deep learning for age estimation (index test), reporting their performance metrically against a reference test (human expert assessment or the actually known age) were included.

PubMed, Google Scholar, Embase, Scopus, ArXiv, medRxiv, and IEEE databases were searched on 24th July 2023, and the search was updated in June 2024.

Out of 2,441 studies, 42 were selected for inclusion. Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. Meta-analysis was restricted to studies (n = 9) that reported the error of the model against the reference test in years.

Thirteen studies demonstrated a low risk of bias, while the majority showed unclear or high risk of bias. Accuracy for classifying individuals into age brackets emerged as the most common metric, with accuracy spanning from 27 % to 100 %. Pooled mean absolute error was 1.75 (95 % CI: 0.96 - 2.55) years CONCLUSION: The performance of deep learning for age estimation from panoramics varied significantly between studies. The mean absolute error, at 1.75 years, however, indicates the promises of deep learning for this purpose.

This systematic review and meta-analysis demonstrated the potential of deep learning as an adjunct diagnostic tool for age estimation, showing that, in mean, the absolute error of deep learning was only 1.75 years. However, several methodological limitations identified herein necessitate further investigation before widespread clinical implementation can be considered.