ACCURACY OF IMAGE-BASED DEEP LEARNING IN ORAL SQUAMOUS CELL CARCINOMA DIAGNOSIS: A SYSTEMATIC REVIEW AND META-ANALYSIS.

The potential of image-based deep learning (DL) in the diagnosis of oral squamous cell carcinoma has been investigated recently. This review aims to evaluate its effectiveness to provide an evidence base for future development. Cochrane Library, Embase, PubMed, and Web of Science were systematically searched up to June 3, 2025. The QUADAS-2 tool was utilized for the risk of bias assessment. Subgroup analyses were conducted by the classification type (binary classification and multiclass classification) and image type (histopathological images, oral photographs, and Raman spectral images). Fifty-two studies were included. For binary classification, DL demonstrated strong performance. Four studies developed DL based on Raman spectral images, yielding a pooled sensitivity of 0.99 (95% CI 0.98-0.99) and a specificity of 0.98 (95% CI 0.96-0.99), with an area under the summary receiver operating characteristic value (AUC) of 0.99 (95% CI 0.20-1.00). Eight studies developed DL based on oral photographs, yielding a pooled sensitivity of 0.97 (95% CI 0.91-0.99) and a specificity of 0.93 (95% CI 0.85-0.97), with an AUC of 0.99 (95% CI 0.85-1.00). DL was developed based on histopathological images in 31 studies, which had a pooled sensitivity of 0.98 (95% CI 0.97-0.98) and a specificity of 0.97 (95% CI 0.96-0.98), with an AUC of 0.99 (95% CI 0.50-1.00). Image-based DL has demonstrated promising diagnostic performance and advantages across multiple imaging modalities including histopathological images, oral photographs, and Raman spectral images, suggesting that it is a promising direction for the development of intelligent diagnostic tools for oral squamous cell carcinoma diagnosis. DL may become an effective adjunct to diagnosis.