Abstract
The proposed research work describes an AIassisted medical chatbot system incorporating a topic-related large language model, convolutional neural networks to analyze medical images, and an automatic appointment system. The proposed system utilizes the Meditron medical language model to obtain patient symptoms and determine the level of urgency through intuitive chat interactions. A convolutional neural network, trained with$\mathbf{1 5, 0 0 0}$marked medical images of$\mathbf{1 2}$typical symptoms, enables visual symptoms to be determined with an average accuracy of 94.3 % on the hold-out set. The appointment system incorporates a rule-based priority algorithm, which takes into account symptoms, specialties, and availability of consulting physicians. The pilot test conducted with 250 users confirmed that average waiting time in consultations was reduced by 47 %, with response times of less than 5 seconds even in cases of simultaneous interaction of numerous users. The study confirms the effectiveness of combining topic-related language models and deep learning algorithms in implementing preliminary patient screening in telemedicine systems.