Deep Learning-based Microscopy Imaging System for Antimicrobial Susceptibility Test

Abstract

The current antimicrobial susceptibility testing (AST) method usually takes a few days and labor-intensive which will delay initial treatment decisions in the early stages of bacterial infections The severe bacterial infection is very likely to cause sepsis and the mortality rate of sepsis will increase by 7 6% per hour when effective treatments are delayed It shows that there is an urgent need for a rapid solution to help doctors prescribe antibiotics more effectively Previous research shows that that the morphology of bacteria will change under beta-lactam antibiotics treatment From the characteristic of bacteria we propose a rapid AST system for Gram-negative bacteria using deep learning algorithms to identify different bacterial morphology changes Moreover using a machine learning method to do the classification automatically determines the minimum inhibitory concentration (MIC) bacteria at different concentration of antibiotics The system uses the Escherichia coli (ATCC 25922) to conduct preliminary study of cefazolin ceftazidime and cefepime treatments The MIC can be determined after 2 hours The accuracy can reach 95% Our system integrates the microscopic imaging system and software which can reduce the detection time from three to four days to 4-5 hours with simple manual operations procedures

Date of Award	2020
Original language	English
Supervisor	Hsien-Chang Chang (Supervisor)