Table 1. Important features of studies included in the systematic review.

Author with year	Objective	Modality	Results	Conclusion
Gomes RFT et al., 2023 [4]	To execute the training as well as the validation of a CNN-based model that automatically categorizes six clinically representative types of oral ulcers photos.	The CNN model InceptionV3	The diagnosis of oral ulcers produced significant results utilizing an InceptionV3-based framework. The authors achieved over seventy-one percent accurate predictions in each of the six lesion categories after hyperparameter optimization. In the collection of data, the categorization had an average reliability of 95.09%.	An AI framework for the automatic categorization of oral ulcers from oral clinical photographs was developed, according to the authors, and it performed satisfactorily.
Guo J. et al., 2022 [5]	To suggest a modified residual network algorithm for instantaneous oral ulcer picture classification.	RN (Residual Network framework)	The findings from the experiment demonstrate that when mouth ulcers are identified and classified in real time, the method developed by the authors surpasses those traditional classification AI networks.	The authors' method works better than the current CNN image categorization techniques and has a fair level of precision in classification. Despite being useful for identifying oral ulcers, the suggested technique needs a broader set of data for validation and training purposes before being used in clinical settings.
Fu Q et al., 2020 [6]	To create a deep learning method that can quickly, non-invasively, cheaply, and simply detect people who have ulceropriliferative growth in the oral cavity corresponding to OSCC from clinical photographs.	DL CNN	Additionally, the approach greatly outperformed both ordinary medical students and ordinary non-medical students in terms of performance.	Automated OCSCC identification using a DL-based technique is a quick, harmless, affordable, and practical approach that has the features to be used as a clinical tool for quick evaluation, earlier identification, and evaluating the effectiveness of cancer treatment.
Zhou M et al., 2023 [7]	To assess CNNs for computerized categorization and identification of RAU, frequently occurring disorders of oral mucosa, and healthy oral mucosa in clinical images.	CNN models ResNet50 model YOLOV5 model	The Pretrained ResNet50 network performed exceptionally well in the process of categorizing, with an accuracy of 92.86 percent. The YOLOV5 model that had been developed had the greatest results, with an accuracy of 98.70 percent.	The categorization and identification of RAU lesions through the use of non-intrusive oral pictures using the previously developed ResNet50 and YOLOV5 algorithms were demonstrated to have better accuracy as well as adequate potential, which may be helpful in clinical practice.
Dar-Odeh NS et al., 2010 [8]	To build and improve an artificial neural network that can foretell the incidence of RAU using a collection of suitable input data.	ANN	The best reliable projections for the likelihood of the presence or absence of RAU were made by the optimized neural network.	The authors discovered variables associated with RAU that might be used as input information to build ANNs that anticipate RAU.
Speight PM et al., 1995 [9]	To assess a neural network's capacity to anticipate the possibility that a person would have a cancerous or potentially cancerous mouth lesion according to knowledge about their risk factors.	Neural network	In general, the dentists' accuracy and precision were 0.74 and 0.99, vs. 0.80 and 0.77 for the artificial neural network.	The aforementioned neural network could potentially be useful for recognizing people with an elevated likelihood of cancer of the oral cavity or precancer of the oral cavity to undergo additional clinical assessment or health awareness training in light of the potential expenses associated with running a screening strategy.
van Staveren HJ et al., 2000 [10]	An alternate approach to categorization for the autofluorescence spectroscopy of oral eryhtroplakia, which may indicate the degree of cellular dysplasia, was assessed to see how well it performed.	Pre-scaled spectra of artificial neural networks.	A neural network could identify between tissue and ulcers with significant accuracy.	The performance of categorizing whether erythroplakia is normal through AI was acceptable.
Wang CY et al., 2003 [11]	This study investigated the likelihood of separating human mouth precancerous as well as cancerous lesions from healthy or normal mouth tissue employing an ANN approach.	PLS-ANN algorithm	The PLS-ANN classification system had an average sensitivity of 81 percent, an accuracy of 96%, and a favorable prognostic value of 88% for separating precancerous tissues and cancerous tissues from normal tissues.	For in vivo identification of OSF in addition to oral precancerous and cancerous lesions, the PLS-ANN classification technique is helpful.
Paul RR et al., 2005 [12]	To provide a unique oral submucous fibrosis grade identification approach that utilizes artificial neural networks.	Artificial neural network	After receiving the photograph as an input, the network that had been trained was capable of classifying healthy stages and oral precancer conditions.	The outcomes of this technique's testing were encouraging, and they point to the possibility of using it for staging, recognizing OSF, and extending it to different settings with more refinement.
Nayak GS et al 2006 [13]	On a comparable set of spectrum data, spectral evaluation and categorization for differentiating between malignant, premalignant, and normal situations were carried out employing principal component analysis (PCA) and ANN individually.	PCA ANN	Both categorization algorithms' specificity as well as sensitivity were assessed. For the data set under consideration, they were 100 percent and 92.9 percent in the context of PCA and 100 and 96.5 percent in the scenario of ANN, respectively.	ANN may help distinguish between cancerous conditions and precancerous conditions.
Kim JS et al., 2022 [14]	On the basis of pictures of the oral mucosa, the effectiveness of AI-related identification of oral malignancy conditions was assessed.	AI, ML, ANN	AI-assisted computerized recognition of oral malignancy conditions could serve as a quick, harmless diagnostic aid that may give instantaneous conclusions on the evaluation for the diagnosis of cancer of the mouth.	A practical approach for the timely identification of pathogenic lesions may be created using this AI technique.
Duran-Sierra E et al., 2021 [15]	This research explored the possibility of using maFLIM-derived autofluorescence indicators in ML frameworks to distinguish between normal oral tissue and malignant oral tissue.	ML, biomarkers of maFLIM-derived autofluorescence	In the present investigation, autofluorescence indicators that are frequently employed in ML frameworks can be used to identify malignant tissue of the oral cavity.	Widefield maFLIM endoscopy, in association with AI, has the ability to automatically recognize cancerous conditions of the oral cavity.
Noyan MA et al., 2020 [16]	To create the TzanckNet DL framework, which can recognize cells within the Tzanck smear.	TzanckNet model DL	TzanckNet generated 2154 estimations for 359 pictures and six different kinds of cells. Precision was 97.3 percent, responsiveness was 83.7 percent, and correctness was 94.3 percent.	The findings demonstrate that TzanckNet possesses the capability to reduce the expertise requirement for using this procedure, thereby increasing the number of its users and enhancing the overall health of patients.
Cai D. et al., 2021 [17]	To build a DL-based screening tool for autoimmune blistering diseases using a unique AI technique.	CNN Transfer learning	The most recent version maintains a level of diagnostic precision for more prevalent skin malignancies that is almost on the same level as the accuracy of dermatologists in diagnosing them.	This method has the ability to aid in the medical diagnosis of AIBDs, which is attributed to the prediction framework's effectiveness despite having inadequate training information.
Dubey S. et al., 2023 [18]	To conduct a comparison study and recommend an AI technique for PV diagnosis in the early phases of skin degeneration.	CNN	A comparison of the results using CNN was carried out with 78.7 percent precision.	The effectiveness of the suggested AI framework for detecting PV was established.
Yu K. et al. 2020 [19]	To review research on ML's use in psoriasis therapy and assessment and to talk about the prospects and obstacles for new developments.	Machine learning	ML has a lot of ability to help in the diagnosis and treatment of psoriasis. The interpretation of medical photographs, the foretelling of consequences, and the development of treatments are currently hot areas in ML studies regarding psoriasis.	Dermatologists would benefit from knowing more about ML and how it might enhance evaluations and decisions, with the goal of allowing patients to profit the most from ML breakthroughs.
Achararit et al., 2020 [20]	To utilize a CNN framework of AI to identify the lichen planus of the oral cavity (OLP) in histopathologically confirmed clinical OLP.	CNN model and Xception framework	All of the chosen CNN models were capable of using images to identify OLP abnormalities. In regard to both general precision and F1 rating, the model built using Xception performed much better than the other approaches.	As demonstrated by their experiment, CNN-based algorithms are capable of achieving an estimated precision of 82 percent to eighty-eight percent. Reliability and F1-score were best achieved with the Xception algorithm.
He X et al., 2020 [21]	To create and assess an AI algorithm for the diagnosis of bullous pemphigoid (BP) and PV.	ML, AI	The AI-based diagnostic structure, which was built on clinical imaging and clinical information, identified BP and PV at a level comparable to dental expert standards.	Dermatologists in rural locations may find this AI-based diagnostic paradigm useful in the early detection of each of these disorders.
Surodina S. et al. [22]	By discovering indicators of HSV infection along with choosing just a handful of pertinent questions that could be asked with new register members to ascertain their probability of HSV risk for infection, this research sought to enhance data collection for an everyday life HSV database.	Random forest algorithm	The model chose fewer items that accurately predicted the likelihood of being infected with HSV with significant high ratings and accurately remembered the danger for HSV-1 as well as HSV-2 data sets.	In an everyday evidence database, this ML system can be utilized to gather pertinent lifestyle information and determine every individual's degree of likelihood of contracting HSV infection.
Natarajan R. et al. [23]	To use and investigate DL techniques based on AI for herpes simplex diagnosis.	Deep learning algorithms, DenseNet, ResNet	DenseNet outperformed ResNet as well as Inception in the particular assignment with a greater level of precision.	The DL approach can be a helpful tool for diagnosing HSV, particularly in primary care settings without access to the necessary laboratory equipment or skilled personnel.
Nowell WB, 2021 [24]	To elaborate on the procedure for creating and putting into use an AI-based approach to lay the framework for creating a patient database for those who are in danger of contracting or already have HSV.	DL	The number of inquiries required to obtain a high degree of precision when forecasting HSV infection was optimized by the researchers employing this technique.	The authors place their novel approach in the broader perspective of both the difficulties in developing a patient register for a stigmatized ailment and the chances to establish new registries by utilizing publicly accessible. data sets.
Idrees M. et al., 2021 [25]	To assess AI's contribution to the diagnosis of OLP.	ML, AI	The proposed machine-learning method was reliably capable of detecting OLP cases based on the number of inflammatory cells and the number of mononuclear cells.	Oral pathologists can more effectively diagnose OLP using aspects of its development thanks to AI, which has demonstrated encouraging results and offers a robust method.
Keser G. et al., 2023 [26]	To create a DL method to recognize OLP lesions from picture photos.	AI, Google Inception V3, Tensorflow, and DL	All experimental photos were correctly classified as represented by an AI framework, with an accuracy of one hundred percent.	The initial outcomes demonstrate that AI possesses the capability to address this important topic.