Neuromorphic Solution for Bacteria Detection in Blood

This technical report presents a neuromorphic solution aimed at enhancing the performance of electronic noses (e-noses) for detecting bacteria in blood samples. The study focuses on implementing a hardware-based, low-power neuromorphic approach that integrates a pattern recognition engine for real-time classification of various bacteria species. The research utilizes a dataset collected as part of the Mednose project, which involved identifying different bacteria species by detecting volatile organic compounds (VOCs) released during bacterial incubation. The methodology includes pre-processing of data samples and the use of two encoding techniques, address event representation for olfaction (AERO) and Step Forward (SF), to convert continuous sensory data into event-based formats. The classification performance of the proposed spiking neural network (SNN)-based classifier is evaluated, demonstrating high accuracy rates with reduced data input requirements. The findings indicate that significant improvements in classification efficiency can be achieved through the proposed neuromorphic system, paving the way for practical applications in medical diagnostics.