Diagnostic performance of convolutional neural network-based AI in detecting oral squamous cell carcinoma: a...

To evaluate the diagnostic accuracy of artificial intelligence (AI) based on convolutional neural network (CNN) in diagnosing oral squamous cell carcinoma (OSCC), we carried out this systematic review and meta-analysis.

We searched PubMed, Embase, Web of Science, ProQuest, Cochrane Library, and Scopus to identify relevant articles from database inception to April 2025. Studies assessing the diagnostic accuracy of AI based on CNN to detect OSCC were included in this search. Statistical analyses were performed by using the Meta-Disc (version 1.4) and Stata 18.0 software.

A total of 14 studies with 61,372 samples were included in the analysis. The pooled positive likelihood ratio (PLR) of 13.08 (95% CI 9.21-18.60) and negative likelihood ratio (NLR) of 0.06 (95% CI 0.03-0.10) were observed with a diagnostic odds ratio of 261.58 (95% CI 131.03-522.19) and the area under the curve being 0.98, respectively. The pooled sensitivity and specificity of CNN based AI in detecting OSCC were 0.94 (95% CI 0.89-0.98) and 0.94 (95% CI 0.92-0.97). Heterogeneity was observed (I² > 75%). Subgroup analyses revealed variations in diagnostic performance based on study design, cancer site, statistical method, external validation, and sample size. The Fagan nomogram indicated that when the pre-test probability was set at 20%, the post-test probability could increase to 81%.

In detecting OSCC, CNN-based AI demonstrates a high diagnostic performance. These findings suggest that CNN models, though not yet widely implemented in routine diagnostic workflows, hold strong potential for OSCC detection. However, the current evidence is largely based on retrospective studies with limited sample sizes and methodological variability, and only one study performed external validation. Therefore, larger prospective and multicenter studies are needed before clinical translation.