Ren Cong, Wang Chengfeng, Lin Ren, Bu Jianbo, Yan Junjie, Wu Mengting
Lishui University, Lishui City 323000 Zhejiang Province, China; Lishui Stomatological Hospital, Lishui City 323000 Zhejiang Province, China.
Department of Stomatology, The People's Hospital of Lishui, Lishui City 323000 Zhejiang Province, China.
J Stomatol Oral Maxillofac Surg. 2025 Jun 13:102429. doi: 10.1016/j.jormas.2025.102429.
To evaluate the diagnostic performance of artificial intelligence (AI) in detecting oral squamous cell carcinoma (OSCC) through a systematic review and meta-analysis.
A comprehensive literature search was conducted in PubMed, Scopus, Web of Science, and other databases for studies published from January 2000 to November 2023. Studies that evaluated AI for OSCC diagnosis with sufficient data to calculate diagnostic accuracy were included. The methodological quality was assessed using QUADAS-2. The primary outcomes were pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR). A bivariate random-effects model was used for analysis.
Twenty-four studies comprising 18,574 specimens were included. The pooled sensitivity was 0.95 (95 % CI: 0.90-0.98), and the pooled specificity was 0.95 (95 % CI: 0.91-0.98). The pooled PLR was 2.60 (95 % CI: 1.91-3.28), and the NLR was 0.10 (95 % CI: 0.07-0.17), with a DOR of 26.0 (95 % CI: 12.1-55.9). Significant heterogeneity was observed across studies (I² = 97.5 % for sensitivity and I² = 97.8 % for specificity). Deep learning algorithms demonstrated superior performance compared to conventional machine learning methods.
AI demonstrates high diagnostic accuracy for OSCC detection, suggesting its potential value as an adjunctive diagnostic tool in clinical practice. However, high heterogeneity among studies indicates the need for standardized methodologies and external validation before widespread implementation.
通过系统评价和荟萃分析评估人工智能（AI）在检测口腔鳞状细胞癌（OSCC）中的诊断性能。
在PubMed、Scopus、Web of Science和其他数据库中进行全面的文献检索，以查找2000年1月至2023年11月发表的研究。纳入那些评估AI用于OSCC诊断且有足够数据计算诊断准确性的研究。使用QUADAS-2评估方法学质量。主要结局指标为合并敏感度、特异度、阳性似然比（PLR）、阴性似然比（NLR）和诊断比值比（DOR）。采用双变量随机效应模型进行分析。
纳入24项研究，共18574个标本。合并敏感度为0.95（95%CI：0.90 - 0.98），合并特异度为0.95（95%CI：0.91 - 0.98）。合并PLR为2.60（95%CI：1.91 - 3.28），NLR为0.10（95%CI：0.07 - 0.17），DOR为26.0（95%CI：12.1 - 55.9）。各研究间观察到显著异质性（敏感度的I² = 97.5%，特异度的I² = 97.8%）。深度学习算法表现出优于传统机器学习方法的性能。
AI在OSCC检测中显示出较高的诊断准确性，表明其在临床实践中作为辅助诊断工具的潜在价值。然而，研究间的高度异质性表明在广泛应用前需要标准化方法和外部验证。
