Ready or Not…

Biomedical Raman spectroscopy has developed rapidly in recent years, with numerous studies demonstrating its potential for improving medical diagnostics.

We are working on two big medical fields of application where Raman spectroscopy offers great chances to fulfill currently unmet medical needs. One is in infectious diseases where it can be used to determine the host response (to define viral, fungal, or bacterial infection) and in case of bacterial infection enable the rapid identification of the causing bacteria and its antibiotic resistance profile. The second big field of application is intraoperative spectral histopathology in terms of tumor boundary detection, staging, and grading. To translate Raman spectroscopy into clinics, surgical microscopic or endoscopic Raman spectroscopy and compact point-of-care Raman setups have been developed in conjunction with clinicians.

However, despite these great advances, Raman spectroscopy has not yet established itself in the clinic – and there are several reasons as to why. One is a lack of reliable tools for the automated analysis of Raman spectra. After all, the success of Raman spectroscopy in biomedical diagnostics is inextricably linked to the development of tailored algorithms for evaluating Raman measurement data (for example, spectral data sets and image data) into qualitatively and quantitatively usable information for end users.

Until recently, the main methods used were based on classical machine learning, but now as the number of Raman datasets and biomedical Raman studies increases, the application of deep learning approaches using neural networks is rapidly entering Raman spectroscopy and becoming increasingly important. And with the growth of AI, we could be on the cusp of a powerful synergy that incorporates both AI and deep learning in Raman spectroscopy. This combination could be the missing link necessary to ensure Raman is routinely used for clinical applications. 

Some hurdles remain. As a community, we are not entirely ready to embrace this technology – far from it, in truth. But I am hopeful that we are open to studying the performance of AI in biophotonics and spectroscopic diagnostics. Recently, I have been reading more manuscripts reporting on automated analysis of biophotonics data – for example, ophthalmic OCT datasets using deep learning – which looks promising.

Undoubtedly, AI will be a game changer. In a few years, it will be indispensable in many fields – and certainly digital pathology. Whether this is good or bad remains to be seen, because, as with many things, AI offers not only benefits but also dangers, which justifies the hesitation of the community.

One concern in using AI in general and deep learning approaches for image analysis and Raman spectral analysis is the foundation on which the analysis is based – something that can be elusive. It is often the case that one does not know the origin of an AI algorithms’ decision. Therefore, it is important to approximate and understand AI tools in order to make them interpretable and understandable for humans.

Of course, the application of AI – not only in medicine – always raises the question of liability. The legal foundations must be laid here in the near future, because I am certain that AI will play a major role in our daily lives in the future. 

This spectroscopic-AI synergy is undeniable and could set the foundation to integrate this fruitful combination of Raman spectroscopy and AI into clinical settings. From a technological point of view, the stage is set and clinically applicable – and medical approved Raman equipment is available. There may be hurdles to overcome before AI’s full potential in clinical research can be reached. But the time is ripe to finally start clinical trials with large patient cohorts showing the great possibility of deep learning in terms of automatically interpreting Raman spectra. However, there are still regulatory hurdles to overcome such as compliance with the EU Medical Device Regulation (MDR) 2017/745 in Europe.

This article was originally published on our sister brand, The Analytical Scientist.