Spectroscopy and machine learning used for health screening

German scientists have developed a health screening tool that uses infrared light and machine learning to detect multiple health conditions with just a single blood plasma measurement. Their breakthrough has been published in the journal Cell Reports Medicine.

Infrared spectroscopy — a technique that employs infrared light to analyse the molecular composition of substances — is effectively like giving molecules a fingerprint that can be delivered by a specialised machine called a spectrometer. When applied to complex biofluids like blood plasma, this physico-chemical technique can reveal detailed information about molecular signals, making it a promising tool for medical diagnostics. Yet despite its longstanding use in chemistry and industry, infrared spectroscopy has not been established nor integrated into the canon of medical diagnostics. A team of scientists from the BIRD group Ludwig-Maximilians-Universität München (LMU) and the Max Planck Institute of Quantum Optics (MPQ) sought to tackle this issue.

Having previously established a method for measuring human plasma, the team collaborated with researchers from Helmholtz Zentrum München to pioneer infrared molecular fingerprinting on a naturally diverse population. This involved measuring blood from thousands of individuals in the KORA study, a comprehensive health research project established in Augsburg, Germany. Randomly selected adults were chosen as a representative scenario for a naturally variable population and were recruited for medical examinations and blood donations.

More than 5000 blood plasma samples were measured using Fourier transform infrared (FTIR) spectroscopy, with members of the BIRD team using infrared light to obtain molecular fingerprints from these samples. The team then applied machine learning to analyse the molecular fingerprints and correlated them with medical data. They discovered that these fingerprints contain valuable information that enables rapid health screening via a multitask computer algorithm that is capable of distinguishing between various health states, including abnormal levels of blood lipids, various changes in blood pressure, type 2 diabetes and even pre-diabetes, which is often undetected. Interestingly, the algorithm could single out individuals who were healthy and remained healthy over the investigated years.

Traditionally, doctors would need a new test for each disease. However, this new approach doesn’t just pinpoint one condition at a time — it accurately identifies a range of health issues, including complex conditions involving multiple illnesses simultaneously. Moreover, it can predict the development of metabolic syndrome years before symptoms appear, providing a window for interventions.

The team’s study thus lays the groundwork for infrared molecular fingerprinting to become a routine part of health screening, enabling doctors to detect and manage conditions more efficiently. This is especially important for metabolic disorders like cholesterol abnormalities and diabetes, where timely and effective interventions can significantly improve outcomes. As the researchers continue to refine the system and expand its capabilities, they hope to add even more health conditions and their combinations to the diagnostic repertoire, which could lead to personalised health monitoring enabling individuals to catch potential issues long before they become serious.

Image caption: PhD student and first author Tarek Eissa has analysed thousands of molecular fingerprints. Image ©Thorsten Naeser/MPQ/Attoworld