Detecção e medição automática de halos pelo método disco-difusão (Kirby-bauer) utilizando aprendizado profundo e visão computacional

Abstract

Bacterial detection and identification at the pathogenic level is essential for making decisions about corrective actions in cases of contamination, so as not to lead to a worsening of the patient's health status. Phenotypic identification is also important in bacterial detection, and one of its methods is to verify the resistance of bacteria to antimicrobials. A classic methodology for this verification is the disk-diffusion test (Kirby-Bauer), where antimicrobial disks are dispensed onto an agar plate after application of the bacterial inoculum, causing the generation of halos. Halos are clear areas formed around antimicrobial disks attached to the Petri dish, called the growth inhibition zone, this measurement is made by a ruler, a time consuming job and subject to human error. Aiming to improve the disk-diffusion results, this work proposes to automatically detect and measure the diameters of the halos for all disks, from 74 digital images in RGB scale. A RNC was trained to segment the area of the disks and the halos, later computer vision was used to perform the measurements of the halos with the segmented images. The results are compared with real values and predicted by the model in computer vision, presenting values of EQM EMA and R² for the training and test data, being: 0.0816, 0.0440, 0.9053; 0.0895, 0.0533, 0.9114 respectively. In this way, it is concluded that the model is a promising technique to be performed on mobile devices, in order to facilitate the laboratory routine.