Modelos preditivos de hipertensão arterial baseados em inteligência artificial

Abstract

Arterial hypertension (AH) is a chronic, multifactorial condition that affects a significant portion of the global population and is one of the main risk factors for cardiovascular disease. Given the need for early detection strategies, this research aimed to investigate the feasibility of using iridological images and health indicators as tools for predicting AH. To this end, Sharp-Íris, an automated iris scanning application using computer vision techniques, was developed, which allowed for the improvement of image sharpness and the identification of signals in iridological tissue. Subsequently, an empirical study was conducted using image analysis by iridologists, involving hypertensive and non-hypertensive individuals, which supported the modeling of the first Bayesian Network based on visual iris signals. As part of the results, a dataset with iris images from hypertensive and non-hypertensive individuals was also organized, which can be used in future research aimed at improving predictive models based on artificial intelligence applied to healthcare. This dataset represents a significant contribution to the scientific community interested in the intersection of iridology, AI, and early disease diagnosis. In parallel, a scientific database containing publications on arterial hypertension was constructed. From this database, natural language processing (NLP) was used to extract causal terms with the TF-IDF algorithm, enabling the construction of a second Bayesian Network based on risk and health indicators. Validation of the proposal included the participation of experts from the fields of healthcare and iridology, as well as the implementation of PrevHealth, an application that integrates the models developed to support the prediction of arterial hypertension based on the analyzed variables. The multidisciplinary approach, involving healthcare, computer science, artificial intelligence, and iridology, enabled the construction of a robust and innovative predictive model. Although the results are promising, the proposal still requires further studies for large-scale clinical validation. Therefore, it represents a potential alternative to conventional diagnostics, focusing on broad applicability in healthcare services, interpretable artificial intelligence, and clinical decision support.