Non-Cochrane · DTA · PM41535840
Diagnostic performance of convolutional neural network-based AI in detecting oral squamous cell carcinoma: a systematic review and meta-analysis.
Shen Mi, Jiang Zhili, Feng Yankun, et al. · BMC oral health · 2026 · PubMed · doi:10.1186/s12903-025-07543-5
Findings
37 findings
| Population | ||||
|---|---|---|---|---|
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · External validation: No | Specificity 0.95 [0.93, 0.97] GRADE not reported 13 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Study design: Prospective | Sensitivity 1.00 [0.98, 1.00] GRADE not reported 1 study |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Cancer site: Tongue | Specificity 0.91 [0.83, 0.99] GRADE not reported 2 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Year of publication: 2019 or earlier | Sensitivity 0.98 [0.93, 1.00] GRADE not reported 2 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Cancer site: Oral cavity | Sensitivity 0.95 [0.92, 0.99] GRADE not reported 12 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Index test: Spectral analysis techniques | Sensitivity 0.98 [0.96, 1.00] GRADE not reported 3 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Statistical method: Without cross-validation | Specificity 0.93 [0.90, 0.96] GRADE not reported 8 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Year of publication: After 2019 | Specificity 0.94 [0.91, 0.96] GRADE not reported 12 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Sample/images: More than 5000 | Specificity 0.93 [0.89, 0.97] GRADE not reported 5 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Study design: Other (retrospective) | Sensitivity 0.95 [0.91, 0.99] GRADE not reported 13 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Statistical method: Cross-validation | Sensitivity 0.98 [0.96, 1.00] GRADE not reported 6 studies |
| Patients or specimens with oral squamous cell carcinoma diagnosed by histological biopsy | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Q* index of the SROC curve | Q* index 0.9425 GRADE not reported 14 studies · 61372 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Statistical method: Cross-validation | Specificity 0.96 [0.94, 0.99] GRADE not reported 6 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Index test: Other index tests | Sensitivity 0.94 [0.90, 0.99] GRADE not reported 11 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · External validation: Yes | Sensitivity 0.95 [0.82, 1.00] GRADE not reported 1 study |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Cancer site: Oral cavity | Specificity 0.95 [0.93, 0.97] GRADE not reported 12 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Index test: Other index tests | Specificity 0.93 [0.91, 0.96] GRADE not reported 11 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Study design: Other (retrospective) | Specificity 0.94 [0.92, 0.97] GRADE not reported 13 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · External validation: No | Sensitivity 0.96 [0.92, 0.99] GRADE not reported 13 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Year of publication: 2019 or earlier | Specificity 0.96 [0.92, 1.00] GRADE not reported 2 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Specificity 0.94 [0.92, 0.97] Moderate Downgraded: inconsistency 14 studies · 2524 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Study design: Prospective | Specificity 0.95 [0.87, 1.00] GRADE not reported 1 study |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Sample/images: 5000 or fewer | Specificity 0.95 [0.93, 0.98] GRADE not reported 9 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Post-test probability after a negative result (pre-test probability 20%) Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Pre-test probability 20% | Post-test probability (%) 1 GRADE not reported 14 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Statistical method: Without cross-validation | Sensitivity 0.92 [0.86, 0.99] GRADE not reported 8 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled positive likelihood ratio Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Positive likelihood ratio 13.08 [9.21, 18.60] GRADE not reported 14 studies · 61372 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Post-test probability after a positive result (pre-test probability 20%) Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Pre-test probability 20% | Post-test probability (%) 81 GRADE not reported 14 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Sensitivity 0.94 [0.89, 0.98] Moderate Downgraded: inconsistency 14 studies · 3115 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · External validation: Yes | Specificity 0.89 [0.77, 1.00] GRADE not reported 1 study |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled diagnostic odds ratio Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Diagnostic odds ratio 261.58 [131.03, 522.19] GRADE not reported 14 studies · 61372 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled negative likelihood ratio Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Negative likelihood ratio 0.06 [0.03, 0.10] GRADE not reported 14 studies · 61372 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Area under the SROC curve Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) | Area under the curve (SROC) 0.98 GRADE not reported 14 studies · 61372 participants |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Sample/images: More than 5000 | Sensitivity 0.97 [0.93, 1.00] GRADE not reported 5 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Sample/images: 5000 or fewer | Sensitivity 0.95 [0.90, 1.00] GRADE not reported 9 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled specificity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Index test: Spectral analysis techniques | Specificity 0.97 [0.95, 1.00] GRADE not reported 3 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Year of publication: After 2019 | Sensitivity 0.95 [0.91, 0.99] GRADE not reported 12 studies |
| Patients or lesions with histologically confirmed oral squamous cell carcinoma versus non-OSCC | CNN-based artificial intelligence (index test) | Histological biopsy (reference standard) | Pooled sensitivity Diagnostic accuracy of image-based CNN models (photographs, histopathology, Raman spectroscopy, PET/CT, hyperspectral imaging) · Cancer site: Tongue | Sensitivity 0.97 [0.91, 1.00] GRADE not reported 2 studies |
Included primary studies 14
Open in Studies| Study | Year | Also included by | Outcomes it informs here | Trial data |
|---|---|---|---|---|
| Ahmad 2023 Multi-Method Analysis of Histopathological Image for Early Diagnosis of Oral Squamous Cell Carcinoma Using Deep Learning and Hybrid… PMID:37958422 | 2023 | 1 other review | not stated | not extracted yet |
| Fu 2020 A deep learning algorithm for detection of oral cavity squamous cell carcinoma from photographic images: A retrospective study. PMID:33150326 | 2020 | 1 other review | not stated | not extracted yet |
| Jeyaraj 2019 Computer-assisted medical image classification for early diagnosis of oral cancer employing deep learning algorithm. PMID:30603908 | 2019 | 1 other review | not stated | not extracted yet |
| Jubair 2021 A novel lightweight deep convolutional neural network for early detection of oral cancer. PMID:33636041 | 2022 | 1 other review | not stated | not extracted yet |
| Warin 2021 Automatic classification and detection of oral cancer in photographic images using deep learning algorithms. PMID:34358372 | 2021 | 1 other review | not stated | not extracted yet |
| Fujima 2020 Deep learning analysis using FDG-PET to predict treatment outcome in patients with oral cavity squamous cell carcinoma. PMID:32524219 | 2020 | this review only | not stated | not extracted yet |
| Kulkarni 2024 Exploring the efficacy of various CNN architectures in diagnosing oral cancer from squamous cell carcinoma. PMID:39610794 | 2024 | this review only | not stated | not extracted yet |
| Lin 2021 Automatic detection of oral cancer in smartphone-based images using deep learning for early diagnosis. PMID:34453419 | 2021 | this review only | not stated | not extracted yet |
| Nagarajan 2023 A Deep Learning Framework with an Intermediate Layer Using the Swarm Intelligence Optimizer for Diagnosing Oral Squamous Cell Carcinoma. PMID:37998597 | 2023 | this review only | not stated | not extracted yet |
| Nanditha 2021 An Ensemble Deep Neural Network Approach for Oral Cancer Screening DOI:10.3991/ijoe.v17i02.19207 | 2021 | this review only | not stated | not extracted yet |
| Soni 2024 Enhancing oral squamous cell carcinoma detection: a novel approach using improved EfficientNet architecture. PMID:38783295 | 2024 | this review only | not stated | not extracted yet |
| Welikala 2020 Automated Detection and Classification of Oral Lesions Using Deep Learning for Early Detection of Oral Cancer DOI:10.1109/access.2020.3010180 | 2020 | this review only | not stated | not extracted yet |
| Xia 2020 Analysis and classification of oral tongue squamous cell carcinoma based on Raman spectroscopy and convolutional neural networks DOI:10.1080/09500340.2020.1742395 | 2020 | this review only | not stated | not extracted yet |
| Yu 2019 Deep convolutional neural networks for tongue squamous cell carcinoma classification using Raman spectroscopy. PMID:31082525 | 2019 | this review only | not stated | not extracted yet |
Overlap with other reviews
Reviews that include the same primary studies are not independent evidence. Shared counts match studies by PMID, DOI or NCT, and otherwise by author and year (a probable match). Only the non-Cochrane dental reviews have study lists, so overlap with Cochrane reviews is not counted yet.
| Review | Shared studies | Jaccard |
|---|---|---|
| Artificial intelligence for oral cancer diagnosis: a systematic review and meta-analysis of image-based and non-imaging models. (2026) | 5 | 0.14 |
Review question
| Element | Planned (methods) | Found (results) |
|---|---|---|
| Population | Patients with OSCC diagnosed by histological biopsy | Patients/specimens with OSCC (61,372 samples; 3,115 positive and 2,524 negative lesions in the pooled analysis; n=14 studies); oral cavity (n=12) and tongue (n=2) |
| Intervention | Any CNN-based AI modality validated for detecting OSCC | CNN-based AI: photographs (n=6), histopathology (n=4), fiber optic Raman spectroscopy (n=2), PET/CT (n=1), hyperspectral imaging (n=1) |
| Comparison | Histological biopsy (reference standard) | Histological biopsy (reference standard) (n=14) |
| Outcomes | Sensitivity, specificity, PLR, NLR, DOR, SROC AUC (2x2 data) | Sensitivity 0.94, specificity 0.94, PLR 13.08, NLR 0.06, DOR 261.58, AUC 0.98 (n=14) |
| Study designs | Diagnostic accuracy studies with 10 or more subjects and sufficient data for a 2x2 table; English language | Diagnostic accuracy studies: 13 retrospective, 1 prospective (n=14) |
| Timing | Not specified | Not specified |