When most of us come across the term “Artificial Intelligence,” we instantly revert back to Stanley Kubrick’s Space Odyssey, or some variation of it. However, we are far from the dystopian world we think “AI” leads to. From environment analysis of self-driving cars, to the evolution of breast cancer screenings, 2019 has seen the highest investment in AI thus far.
Current methods and complications
When it comes to oral health, one area that subjective clinical judgment can have a life saving outcome is the oral cancer screening. The most critical factor for a positive outcome in oral cancers is early detection. Unfortunately, more than two thirds of all oropharyngeal squamous cell carcinomas (OPSCC) are detected only after they have metastasized. Its not surprising that OPSCC have the highest morbidity all major cancers.
Oral health care providers have the most optimal opportunity to detect OPSCC. The current practice largely involves visual and tactile examination of oral mucosa, and if anything suspicion is present – referral to a specialist for biopsy. But the subjective decision making of clinicians, when considering the features such as color, texture, and consistency has shown that our accuracy varies greatly amongst each other. In one study, dentist choosing between OPSCC and a benign lesion showed 57.8% sensitivity and 53% specificity. Another study showed only 31% specificity.
What can AI do?
What artificial intelligence and its subsets (machine learning and deep learning) can do is analyze and cross- reference vast amounts of data. This data can be in the form of an image, disease risk factor, geography, and signal intensities – combing all data points to produce a risk assessment for OPSCC. To date, two approaches using AI have been undertaken and they show promise.
A smartphone probe
Sankaranarayanan et al. showed that with the use of a low-cost smartphone probe, we could analyze autofluorescence and polarization images and cross reference them with OPSCC risk factors such as habits, signs, pre-morbidities. The outcome of such an analysis aided greatly in decision making of whether a case needed further treatment/consideration. It has shown that this method of triage shows an accuracy of 85%, compared to health care worker detection ranging from 30% to 60%.
Mapping heterogeneity and Margins of OPSCC lesions
Another application of AI has involved using an image-processing app (Heidari et al. 2019) and classifying lesions as normal, dysplastic, or malignant. This allows the health care worker not only to have an aid in deciding what is normal or not, but also what is the right next step. This process “differentiated between healthy versus dysplastic versus malignant tissues with a sensitivity of 87% and a specificity of 83% versus the histopathological gold standard”.