HubDoctor - computational platform to assess the severity of oral problems, scheduling urgency, indication of the most qualified professionals for the case, patient-professional communication and personalized scheduling

Abstract

The results obtained on PIPE FAPESP PHASE I showed great potential for commercialization of the proposed innovation, for they showed that people have little knowledge about oral healthy and search information on the internet to have their questions answered about their problems and the dentists are unhappy and are in search of solutions to achieve quick professional success. The HubDoctor platform offers and innovative and disruptive service, entirely by using a virtual assistant powered by an artificial intelligence (AI) specialized in dentistry (healthbot HubDoctor) which evaluates the risk of oral pathologies, personalized orientation and the ability to indicate professionals with the highest probability to solve the user's problem. In order to make it all possible, it will be necessary to overcome the following technological challenges: 1) is an assistant chatbot specialized in dentistry capable of keep a dialog fluently and consistently with the users and help them solve their oral problems in every region of the country? 2) is the free text dialog structure more efficient than a structured menu for the healtbot to make the proper evaluation of risk of acid erosion and professional indication? 3) can the healthbot collect data to determine the predictive models of risk evaluation and professional indication? 4) do these predictive models have an accuracy similar to that of professional experts in the field to orient the patients in a trustworthy and assertive manner? For the PIPE PHASE II, there were proposed 4 researches with the objective of: 1) evaluate the performance of two architectures (micro services or conventional); 2) validate scientifically the predictive model of the evaluation of risk of acid erosion developed by neural networks and deep learning and confirm its accuracy in comparison with that of a field specialist; 3) scientifically evaluated the predictive model of indication of professionals developed by neural networks and deep learning and confirm its accuracy in comparison with a recommendation given by a field specialist; 4) evaluate the healthbot's conversational performance with users of different social classes and regions of the country. The obtained data will be statistically analyzed using the Kappa coefficient, X2 and/or regression considering alpha = 0.05. With the realization of this project, it is expected to identify the most adequate predictive models for the evaluation of risk of acid erosion and for the indication of the professionals, as well as the best strategy to develop a healthbot capable of helping both the common user and the professionals with trustworthy orientations and scientifically validated. (AU)