Machine Learning and Electronic Noses for Medical Diagnostics

The need for noninvasive, easy-to-use, and inexpensive methods for point-of-care diagnostics of a variety of ailments motivates researchers to develop methods for analyzing complex biological samples, in particular human breath, that could aid in screening and early diagnosis. There are hopes that electronic noses, that is, devices based on arrays of semiselective or nonselective chemical sensors, can fill this niche. Electronic olfaction uses data processing and machine learning to build classification models based on the responses of several sensors in the form of multivariate datasets in order to discriminate between disease and healthy control based on a unique fingerprint. However, the introduction of this technique in clinical settings is limited by methodological issues which can, to some extent, be remedied using artificial intelligence. In this chapter, we provide a brief introduction to the electronic nose technique and outline its applications in medical diagnostics. We also discuss the ways in which data processing and machine learning techniques can be used to facilitate the use of electronic olfaction in the detection of disease.