Deep-learning aided bacteria diagnostics using multimodal optical spectroscopy

Abstract

Microorganism identification in biological samples is a crucial step for defining the pathogen responsible for patient’s sickness. Carrying it out as fast and labor efficient as possible is beneficial for patient’s successful and timely recovery, since it helps to assign the correct antibacterial therapy sooner. The classic method for microorganism identification is biochemical analysis, but this method greatly suffers in terms of time and labor consumption since it requires the microbiologist to consequently observe the reaction of biological sample with an array of biochemical agents, and the process is inherently limited by the rate of the reaction. More modern methods, such as mass-spectrometry usually offer a much faster identification, but they mostly suffer from the need of preliminary sample preparation for higher accuracy of classification and don’t allow the microbiologist to carry out the identification right from the microorganism colonies cultivated in the growth medium on the Petri dish. The capabilities of microorganism identification methods that use the optical response data for the cultures grown on Petri dish were tested in this work, and deep-learning algorithms, capable of extracting complex patterns from the input data, were employed. We will discuss the proposed method for intelligent combination of optical data of different modalities for prediction of object type in a single classification pipeline, leveraging the predictive ability advantages of different optical methods.

Speaker

Oleg Pavlov
Faculty of Physics, M.V. Lomonosov Moscow State University, Russia
Russia

Discussion

Ask question