Uma revisão sistemática da literatura sobre o uso de aprendizado de máquina para o auxílio ao diagnóstico de exames de manometria anorretal e esofágica

Abstract

Manometry exams are a fast and safe way to detect and diagnose functional disorders of the digestive tract with high accuracy. Because of this, these exams are considered useful in the field of gastroenterology, and there is interest in the clinical field in improving the procedure of these exams with new technologies- in particular machine learning. This work seeks to contribute to this academic effort through a systematic review of the existing literature. The PubMed, Scopus and IEEE repositories were used as sources. A total of 16 articles were selected and analyzed, each detailing the implementation of machine learning tools to aid in the diagnosis of esophageal and anorectal manometry exams. It was concluded that systems that use pre-trained convolutional neural networks are the current state of the art, capable of generating complete diagnoses. However, to train these networks, a dataset with thousands of images is required. If one does not have access to a dataset of this size, it is recommended to use models such as multilayer perceptrons, quantization of learning vectors, support vector machines, k-nearest neighbor, logistic regression, and random forests. These systems are not capable of generating complete diagnoses, but they can identify useful information which can aid in evaluating examssuch as the detection of normal and/or abnormal movements and the assessment of integrated relaxation pressure.