Desenvolvimento de uma modelagem de inteligência artificial para facilitar a comunicação de pessoas surdas nos serviços de saúde

Abstract

In the context of healthcare services, although legislation exists to ensure inclusive, accessible, and specialized care, several communication barriers still hinder full access for deaf people, as their native language is Libras, which is still little used and understood by the majority of the population. This study's overall objective was to evaluate the communication tools used by deaf people to interact in healthcare services. Specifically, we sought to identify the communication methods employed by deaf people in these contexts; assess the impact of communication barriers on healthcare service uptake; investigate the use of apps and software as communication resources; and develop an artificial intelligence model capable of converting some Libras signs into Portuguese, with a focus on healthcare. This is an action research project, with a theoretical and practical approach, developed in three phases: exploratory, planning, and action. The main sample consisted of deaf individuals of both sexes, aged 18 and over. In the exploratory phase, a structured questionnaire focused on communication barriers and app use by deaf individuals. In the planning phase, the main apps reported by participants were identified, their limitations analyzed, and, based on this, a model for a Libras-Portuguese translator with healthcare-related signs was developed and trained. Finally, in the action phase, the model was validated using the following metrics: precision, recall, F1-score, error rate, accuracy, and macro averages. The results confirmed the existence of significant communication barriers between deaf individuals and healthcare professionals. Although both seek alternative ways to communicate, these strategies are still insufficient, leading many deaf individuals to rely on companions. Even with this support, many reported poor understanding of medical instructions, which contributes to low adherence to healthcare services. Among the most cited apps, ICOM and Hand Talk stood out, recognized as tools with potential but still have limitations, the main one being Libras-to-Portuguese translation, as evidenced by Hand Talk. This gap was the focus of the modeling developed in this study. Although the model's results were lower than expected (less than 40%), the technology is considered promising and has the potential to significantly contribute to communication between deaf people and healthcare professionals.