Robodoc: a conversational-AI based app for medical conversations

Abstract

AbstractArtificial Intelligence (AI) evolved in trends. Currently, the trend is Conversational Artificial Intelligence (CAI). Those models of AI are focused on text-related tasks, and their commonest applications are chatbots. On this paper, we explore a smart chatbot using the Large Language Models (LLMs) from openAI. I have used a tool called Teachable Machine (TM) from Google to apply transfer learning and create image-based models. I have built two image-based model: for X-ray and for OCT. The model of X-ray is able to detect viral and bacterial pneumonia, whereas the Optical coherence tomography (OCT) model can detect Drusen, Choroidal Neovascularization (CNV) and (Diabetic Macular Edema DME) conditions on the patient’s eyes image. I have also used TensorFlow.js from Google to create a diabetes detection model. All those models are integrated into a chatbot, that according to the message entered by a user, is able to use the models intelligently. Our results show a good integration between the models and the chatbot, with slight deviations from the expected behaviors. For the OCT model, we have also tested a stub function for medical appointments done by the bot, based on how serious is the patient condition. The future of artificial intelligence are public APIs, as I have shown that a complex model can be built, without a complex research infrastructure, and with low costs. Bioinformatics may have gained a new supporter towards more friendly interfaces on bioinformatics.