Evaluation of Artificial Intelligence in Diagnosis and Risk Assessment of Oral Potentially Malignant Disorders

Overview

Oral potentially malignant disorders (OPMDs) are mucosal lesions that carry a risk of malignant transformation into oral cancer. Unfortunately, a general lack of knowledge and awareness of OPMDs is common among general dental practitioners. While thorough clinical examinations coupled with biopsy can identify most OPMDs, the absence of reliable non-invasive diagnostic tools and standardized risk stratification often delays early diagnosis and treatment of oral squamous cell carcinoma (OSCC).Early detection of suspicious oral lesions is crucial for reducing OSCC-related mortality and improving patient outcomes. Histopathological assessment of biopsied tissue remains the gold standard for diagnosis. However, since biopsy is invasive and may be associated with patient discomfort; numerous noninvasive diagnostic technologies have emerged to enhance the detection and diagnosis of oral mucosal lesions.Toluidine blue (TB) staining is one such adjunctive tool, where the degree of color retention aids in lesion characterization. Dark blue staining is considered positive for lesions highly suspicious for malignancy; light blue retention is considered positive for premalignant lesions pending histopathological confirmation, while lesions showing no stain retention are classified as negative.Exfoliative cytology represents another non-invasive diagnostic approach, wherein cells obtained via brushing the oral mucosa are spread on a slide for cytological evaluation. This technique, widely accepted and increasingly utilized, has proven valuable for early cancer detection. Notably, confocal microscopy has demonstrated high sensitivity and specificity (93%) in detecting malignant cells in exfoliative cytology specimens. Currently, TB staining and confocal microscopy remain the most commonly utilized non-invasive screening techniques in clinical practice.In recent years, artificial intelligence (AI) applications have shown remarkable promise in oncology, achieving high diagnostic accuracy across various cancer types. Deep learning models, in particular, offer exceptional performance, suggesting that AI-based solutions may be feasible for widespread community screening programs following further validation. In many cases, AI models have produced diagnostic outcomes that match or surpass those of experienced pathologists. Moreover, the combined application of AI with expert human evaluation has been shown to reduce diagnostic errors and improve diagnostic precision, particularly for poorly differentiated tumors and rare cases.Several studies have been done using different AI Models and revealed a promising application of AI in diagnosing OPMDs and cancers in different body sites.

Oral potentially malignant disorders (OPMDs) are described as the mucosal lesions that have the potential to be oral cancer. It is consisted of oral leukoplakia (OLK), oral lichen planus (OLP), oral erythroplakia (OEK), discoid lupus erythematosus, proliferative verrucous leukoplakia, candida leukoplakia, reverse smoker's palate, verrucous hyperplasia, dyskeratosis congenita, actinic cheilosis, keratoacanthoma, and oral submucous fibrosis. Up to 5% prevalence was reported in the literature for OPMDs and common localizations were described as tongue, the floor of the mouth, and gingiva. The malignant transformation rate of OEK, OLK, and OLP was estimated approximately 14.3%-50%, 0.13%-17.5%, and 0.4%-6.5%, respectively.Since many oral squamous cell carcinomas (OSCCs) develop from OPMDs, clinicians must distinguish those lesions with thorough diagnosis and management to prevent malignant transformation. Lack of knowledge and awareness about OPMDs are common in the general public and studies demonstrated that general dental practitioners are not fully informed/ prepared for those entities . Diagnosing OPMDs as definable diseases is also challenging due to the numerous varieties, various forms, and overlapping features. However, studies have found that when an OPMD changes to a nonhomogeneous presentation, it is more likely to be considered as an adverse progression, in other words, nonhomogeneous lesions have a greater risk of malignant transformation as against homogeneous lesions.Oral tissue biopsy and histopathological analysis are often considered as the gold standard for cancer risk assessment of OPMDs. However, since biopsy is an invasive assay, it may not be suitable for monitoring the chronic development of OPMDs when compared to non-invasive detection techniques. Another disadvantage that the biopsy requires, on average, a day and half for a report. Although, thorough clinical examinations with the help of biopsy may reveal most of the OPMDs and OCs, other diagnostic methods such as vital staining, microfluidics, salivary diagnostics, and cytopathology platforms could be utilized. Toluidine blue (TB) stain is a basic metachromatic dye of thiazine group that shows affinity for the perinuclear cristernae of DNA and RNA with greater penetration and temporary retention of the dye in the intercellular spaces of rapidly dividing cells in-vivo RNA . It has high sensitivity (73.9%) and low specificity (30%). Reports have concluded that toluidine blue retention in high risk OPMDs and high-risk molecular clones, even in lesions with minimal or no dysplasia have documented .While, acridine orange (AO) is a histochemicalfluorochrome with a selective affinity for nucleic acids. At a concentration of 0.01% and a pH of 6, recommended by Von Bertalanffy, the DNA fluoresces yellow to whitish green and the RNA red. AO is a low molecular weight, weakly basic dye that easily penetrates cell membranes. AO has ametachromatic properties and upon excitation with blue light, (488 nm) it emits green fluorescence. Exfoliative cytology is a diagnostic procedure which has been generally accepted for early diagnosis of cancer.Confocal laser scanning microscopy is an advanced microscopic imaging technique which was found to have good sensitivity in the identification of malignant cells in exfoliative cytology. The advantages offered by this technique are the rapidity of processing and screening the specimen and addingobjectivity to the process of clinical diagnosis. Acridine orange-stained confocal microscopic images has showed good sensitivity and specificity (93%) for detection of OPMDs.Artificial intelligence (AI) is a branch of computer science which can be defined as the ability for a computer to mimic the cognitive abilities of a human being. AI corresponds to a large array of techniques. Among them, deep learning is a potential disruptive technology that attempts to model high-level abstractions in medical images to determine diagnostic meaning. Deep learning, specifically as implemented using convolutional neural networks (CNNs), has become a conventional technique for classifying, detecting, and segmenting the objects in medical images. Artificial Intelligence (AI) and Machine Learning (ML) have gained extensive attention in dentistry to achieve cognitive functions of clinicians such as differentiation, problem-solving, and learning. Nowadays, computer-aided diagnosis systems (CAD) which were powered by convolutional neural networks (CNN) were able to detect and classify some cancerous lesions. Some of recent studies regarding using of AI in oral cancer early detection, diagnosis, and treatment outcome concluded that the Machine learning technique has the potential to help in oral cancer screening and diagnosis based on the datasets. Also the deep and conventional learning algorithms are successful in predicting the malignant transformation of oral leukoplakia and oral lichenoid lesions.The Deep CNNs can be an effective method to build vision devices with limited memory capacity for the diagnosis of oral cancer, while Faster R-CNN models have the highest detection performance, with an AUC of 74.34% and potential for classification with high sensitivity and specificity ( 100% and 90%) respectively For the CNN-based classification model. So this study is aiming to have a deep learning algorithm model that is capable for risk assessment in addition to detection and classification of OPMDs with high accuracy, sensitivity and specificitycompared to experts.