Abstract:

Dentistry clinical training challenges have been reported in the transition from pre-clinical to clinical environment. This transition has provoked several adaptations to the dental school's curriculum in the US. Nowadays, with the integration of digital tools and artificial intelligence in the health system, the challenge is integrating these resources into the educational environment and clinical setting to take advantage of them and facilitate the translation of theoretical knowledge to applicability in real time while seeing patients. The American Dental Association (ADA) has recently approved two new specialties in dentistry; these are the disciplines of Orofacial Pain (OFP) and Oral Medicine (OM). These two disciplines are more commonly established in healthcare centers for education as hospitals and Universities therefore, the majority of the patients with OFP and OM conditions are seen by trainee providers (clinical residents) under the supervision of senior clinicians. The significant subjective component and multiple symptom presentations of the diseases in OFP and OM present a big diagnostic challenge for dental residents. Although some of the suspected oral diseases, disorders, and dysfunctions in the OFP and OM arena are finalized with histopathology or imaging, many do not have biomarkers or disease-defining diagnostic tests. Providers and clinical residents must often rely on a combination of subjective and objective observations to diagnose a patient who might have multiple complaints, which represents a challenge for novice clinicians, specifically first-year residents. The clinical observations in these two disciplines are typically recorded as free text, making the information difficult to process for the novice clinician and mine computationally. To address these issues, we elected to create a customized note-taking system called Smart Note (SN) for the Orofacial Pain (OFP) and Oral Medicine (OM) Center at the University of Southern California (USC). Usually, a regular patient encounter in the OFPOM center creates approximately 100 observations. The SN has available 900 possible observations to choose from and 215 diagnoses integrated. The design of the SN provides branch points for checkbox access and displays the probabilities of the currently most likely diagnoses to support the clinician's decisions. Those probabilities were computed by the Naïve Bayesian inference algorithm trained on 1195 de-identified encounters recorded in our dataset. We used 211 new cases for validation. While some rarer diagnoses are still problematic (low accuracy), we have created a structured and accurate note-taking system that shows a promising ability to predict the expert’s true positive working diagnosis, supporting the clinical decision-making process of novice clinicians.

Keywords:

Artificial intelligence, diagnosis, clinical training, dental residents, orofacial pain, oral medicine.