Breath VOC analysis and machine learning approaches for disease screening: a review

Abstract

Early disease detection is often correlated with a reduction in mortality rate and improved prognosis. Currently, techniques like biopsy and imaging that are used to screen chronic diseases are invasive, costly or inaccessible to a large population. Thus, a non-invasive disease screening technology is the need of the hour. Existing non-invasive methods like gas chromatography-mass spectrometry, selected-ion flow-tube mass spectrometry, and proton transfer reaction-mass-spectrometry are expensive. These techniques necessitate experienced operators, making them unsuitable for a large population. Various non-invasive sources are available for disease detection, of which exhaled breath is preferred as it contains different volatile organic compounds (VOCs) that reflect the biochemical reactions in the human body. Disease screening by exhaled breath VOC analysis can revolutionize the healthcare industry. This review focuses on exhaled breath VOC biomarkers for screening various diseases with a particular emphasis on liver diseases and head and neck cancer as examples of diseases related to metabolic disorders and diseases unrelated to metabolic disorders, respectively. Single sensor and sensor array-based (Electronic Nose) approaches for exhaled breath VOC detection are briefly described, along with the machine learning techniques used for pattern recognition.

Progress and challenges of developing volatile metabolites from exhaled breath as a biomarker platform

“…Another approach is to use more complex analytical workflows. Machine learning approaches are increasingly being used to analyze breath VOCs, aiming to identify patterns in how relevant compounds cluster together (Haripriya et al, 2023 ) This can include other clinical data to maximize the usefulness, and interpretation of breath test results, and has been utilized with success previously (Ibrahim et al, 2022 ). Handling highly complex VOC data in breath is key to identifying meaningful differences between study cohorts, especially if the origin of the identified VOCs in the body is known, and the normal ranges in a healthy population are available in a reference database (Fig.…”

“…Gaining a deeper understanding of the networks of interactions of VOCs can help to relate them to underlying physiology, and therefore algorithms can learn what are likely to be false-positive hits. Utilizing the patterns of breath data rather than individual VOCs is similar to how e-nose technology operates (Hao & Xu, 2017 ; Haripriya et al, 2023 ; Seesaard et al, 2012 ; Shlomi et al, 2017 ), with different sensors responding to different chemical groups. However, sensor-based techniques often suffer from a lack of repeatability, as well as technical issues such as drift and sensor faults (Haripriya et al, 2023 ; Le Maout et al, 2018 ).…”

“…Utilizing the patterns of breath data rather than individual VOCs is similar to how e-nose technology operates (Hao & Xu, 2017 ; Haripriya et al, 2023 ; Seesaard et al, 2012 ; Shlomi et al, 2017 ), with different sensors responding to different chemical groups. However, sensor-based techniques often suffer from a lack of repeatability, as well as technical issues such as drift and sensor faults (Haripriya et al, 2023 ; Le Maout et al, 2018 ). The ability to identify specific VOCs rather than just overall patterns makes spotting likely contaminating VOCs that are more likely to be unrelated to underlying biology easier.…”

Artificial intelligence enhanced sensors - enabling technologies to next-generation healthcare and biomedical platform

“…AI algorithms are becoming increasingly important for VOC gas detection because of their benefits for VOC sensors (Haripriya et al 2023 ; Lekha and Suchetha 2021 ; Lotsch et al 2019 ; Mahmood et al 2023 ; Thrift et al 2019 ; Zhou et al 2021 ; Zhou, H. et al 2023a ). First, the use of machine learning algorithms facilitates the automated design of volatile organic compound sensors, thereby eliminating the need for arduous and time-consuming design processes.…”