Electronic nose based bacteria species detection in diabetic foot infection

Abstract

This thesis presents a fundamental study of early bacteria detection using electronic nose. There is a need for early detection of bacterial infection in order to give effective treatment for diabetic foot infection. To date, the clinical method based on sample culture is a standard practise used by microbiologist to detect and classify bacteria species. The cultured samples were taken from debridement of diabetic foot wound can take up to two to three days. Alternatively, identification of causative bacteria from their odours could provide an early and rapid diagnosis and therefore allow initiating appropriate treatment. This research project used an existing sensor technology in the form of an e-nose in conjunction with data processing and classification methods to classify six types of bacteria, common causal organism of diabetic foot infection from their odours. There were three main aims in this research study namely, to identify different bacteria species in different culture media using e-nose, investigate the ability of the e-nose to detect cultured bacteria species in blood agar medium in less than 24 hours and study in-vitro diagnosis of single and poly microbial species targeted for diabetic foot infection using e-nose. Cyranose 320 e-nose device which consist of 32 gas sensor array, measures the changes in resistance of each chemical sensor which can detect and classify bacteria according to their volatile organic compound (VOC). The sniffing process or e-nose measurements were performed immediately after placing the petri dish of bacteria suspension in a special stainless steel container. The odour data were collected and stored as numerical values within data files in the computer system. Once the dataset extracted, various classification experiments were performed. Comparisons were made and conclusions were drawn from the performance of various data analysis and classification methods. The classification methods used in this work include Support Vector Machine (SVM), K Nearest Neighbor (KNN), Linear Discriminant Analysis (LDA) and Probability Neural Network (PNN). 100% accuracy was achieved using all classifiers for identification of bacteria species in three different culture media. The results confirmed that possible to discriminate different bacterial groups on diabetic foot infection regardless of different culture media used for bacteria growth. For early detection of six bacterial species, the best accuracy was 96 %. This was achieved using KNN with k value of 2 and 6 using Euclidean and City block distance. For study in-vitro diagnosis of single and poly microbial species, the best accuracy was up to 90 % for all classifiers. Thus, this fundamental work on the classification of bacteria odours using e-nose can be a „real world‟ application if this technology is successfully developed. The methods and techniques discussed here are one step towards the goal of introducing multi class sensor systems into everyday use. The conclusion of this thesis is that an e-nose can detect and classify different types of bacteria on diabetic foot infection with convincing results which are comparable to the existing standard procedure.