Capucine, how does it feel being a CYIA 2026 Awardee?
I think I'll fully process it over the coming weeks. It feels very satisfying because I spent a lot of weekends working on this project. At the beginning, it was actually just for fun. I simply wanted to do something outside of my PhD work and see where it would lead. And now it has turned into a real project. I can now see its potential clinical impact. It's very fulfilling for me. Honestly, it's like a little dream come true.
Let's talk about the project itself. So you work on analyzing the results of magnetic resonance spectroscopy, right? Which non-invasively examines the chemical composition of brain. What is it currently capable of measuring and detecting?
MRS (Magnetic Resonance Spectroscopy) is a technique that allows us to measure metabolites in tissues. Our research mainly focuses on brain tissue, particularly in the context of brain tumors, although the method can also be applied to many other tissues throughout the body. Using MRS, we can detect biomarkers associated with different cell populations, such as glial cells, and with processes such as neuroinflammation. These biomarkers provide valuable information about the metabolic state of the tissue and help us assess how healthy or abnormal it is. For example, we can measure metabolites such as N-acetylaspartate (NAA), which is considered a marker of neuronal integrity because it is found almost exclusively in neurons. We can also measure choline-containing compounds, which are associated with cell-membrane turnover and are predominantly, though not exclusively, of glial origin, as well as creatine, one of the most commonly measured metabolites, which reflects cellular energy metabolism.
These are the metabolites that can usually be measured quite reliably. However, our work also focuses on a number of less abundant metabolites that are more challenging to detect but that provide highly valuable information about specific tissue alterations. These metabolites can be particularly useful for identifying certain diseases or for distinguishing between different pathological conditions. Another important metabolite is myo-inositol, which is considered a putative marker of glial cells, especially astrocytes, one of the main types of glial cells. In fact, there are many metabolites that can be detected using MRS, each providing different insights into tissue metabolism. In my own PhD research, I have mainly focused on 2-hydroxyglutarate (2-HG). This metabolite accumulates specifically in IDH-mutant gliomas, a subtype of adult-type diffuse glioma; gliomas are the most frequent primary malignant brain tumors in adults. Detecting 2-HG non-invasively is one of the most promising clinical applications of magnetic resonance spectroscopy and illustrates its potential impact on the diagnosis and management of brain tumors.
Why isn't this method widely used in clinical practice yet?
There are several reasons. There is a lack of expertise, and few people know how to acquire this kind of data. Before a magnetic resonance spectroscopy acquisition, you need to place a voxel—essentially the three-dimensional equivalent of a pixel—very precisely within the region of the brain you want to study. Another challenge is that MRS data contain many unwanted signals that can make it difficult to detect the metabolites you're actually interested in. For example, spectra can be contaminated by lipid signals, which in the brain arise mainly from extracranial fat and, patients with brain tumors, fromnecrotic tissue; edema is also a source of degradation. Spectral overlap of many metabolites at once can also hinder reliable detection. All of these components can contaminate the spectrum and interfere with the signal you want to measure. Although there are very advanced acquisition sequences designed to minimize these effects and optimize data quality, using them still requires considerable expertise. The second challenge is data processing. Interpreting MRS spectra is complex and typically requires specialized software and expert knowledge. This is exactly where my project comes in.
The goal is that, once the data have been acquired, the entire analysis is performed automatically. Clinicians would no longer need to rely on complex, research-oriented analysis pipelines that often require manual processing and are not easily accessible outside specialized research centers. Instead, the software processes the data automatically and generates a metabolic profile of the tissue, providing detailed information about the chemical composition of the specific brain region being examined. This information can then support the diagnostic process. Ultimately, the aim is to democratize access to MRS in clinical practice. We are becoming increasingly aware of the clinical value of this technique, and I believe that in the near future it will be implemented in many more hospitals and medical centers. That's why I think now is the right time to develop tools that make this technology accessible to everyone, not just specialists in spectroscopy or data processing. Today, existing tools are either research-oriented, single-vendor, single-sequence, locally installed, or non-standardized in their processing. The contribution here is an accessible, standardized, hosted pipeline. Ideally, the outcome of this project will be to make this type of analysis easily accessible to clinicians, giving them the independence to analyze their own data without relying on specialized experts. At the same time, the platform also has an educational value. It allows clinicians to explore their own data, become more familiar with spectroscopy, and better understand the information it provides. I think that's an important way of sharing knowledge about magnetic resonance spectroscopy, which is a very promising technique with great potential for the future.
What are the biggest limitations of the method at the moment?
We can say that the biggest limitation of this current approach is its complexity in adoption: it is difficult for clinicians to use in medical practice and difficult to evaluate the results. There are two bottlenecks, first, the acquisition, and second, the processing and the interpretation. These are the big points we should focus on.
Is there a risk that the results could be overgeneralized?
The challenge in the MR spectroscopy community is that every center has its own processing pipeline. As a result, it's often difficult to compare analyses across centers, because the results are not always reproducible. The processing steps can be completely different from one center to another. This makes it difficult, both in research and in clinical practice, to have a benchmark or a common ground truth against which results can be compared. I'm not saying that my project provides the ground truth, but it is a step towards greater standardization across centers. The idea is to provide a common processing pipeline that allows results to be compared more consistently from one site to another. Of course, the long-term goal is to adapt this pipeline to different MRI systems and different acquisition sequences. There's still a lot of work to do, but the main objective is to support clinical practice by helping clinicians interpret MRS data in a consistent way across different centers, making the results more reliable and easier to interpret. That is one of the main goals of this project as well.
So in the future, could this method also help identify new relationships or biomarkers that we haven't yet associated with specific conditions? I mean, new causalities and new relationships before markers for substances in our brain and diseases…
Yes. The goal is to gather more and more data with time from our own center and other collaborators. The more data we have, the more opportunities we have to analyze them and discover new biomarkers. On the website, for example, a numberof visualanalyses are performed automatically. You can generate heat maps, compare metabolite concentrations acrosspatients, and explore the data in different ways. I think tools like this will make it much easier to identify patterns associated with specific diseases or patient populations. They can also accelerate new discoveries, because the software performs the analyses automatically and identifies associations between metabolites. This makes it easier to uncover broader relationships between different metabolic features and diseases. These analyses are hypothesis-generating, any association they highlight has to be confirmed in an independent, appropriately powered study before it can be interpreted as a biomarker. Looking ahead, another goal is to integrate additional types of information. Right now, the platform mainly focuses on MR spectroscopy, but in the future we would also like to include physiological data, whenever they are available, to support diagnosis and provide a more holistic view of the patient's profile. Ultimately, the idea is to combine different sources of information to improve diagnostic accuracy. Genomic data could also be very valuable. I work with research teams that specialize in genomics, so I can definitely see the potential. However, genomic data are not always readily available. They are expensive to obtain, time-consuming to analyze, and require highly specialized expertise, which is not available in every hospital or clinical setting.
The idea is to have a kind of stepwise approach. MR Spectroscopy would be the starting point, and if additional data are available, they can be integrated to provide a more comprehensive and accurate diagnostic profile. Of course, we would also like to incorporate other MRI modalities that are already widely used in clinical practice. For example, anatomical MRI sequences such as T2-FLAIR provide valuable information about hyperintensities associated with different diseases. In gliomas, for instance, they are very informative, but they can also reveal many other types of abnormalities in brain tissue. There are also other MRI sequences that provide complementary information. Contrast-enhanced images acquired with gadolinium, for example, can highlight differences in tissue behavior and provide additional insights that complement the spectroscopy data. So the long-term goal is to build a modular platform that can integrate different types of information and ultimately support a multimodal analysis, which is generally more informative than relying on a single modality alone.
The challenge is that these additional data are not available for every patient. Acquiring multiple imaging modalities can be expensive and time-consuming, and in many clinical settings it is simply not feasible to collect all of them routinely. That's why the analysis should adapt to whatever data are available, and I think spectroscopy provides a very good baseline for this approach. With a platform like this, it will become much easier to analyze data, explore relationships between different types of information, and perhaps even identify patterns that we hadn't previously considered. That's often how research works—sometimes the most interesting discoveries happen unexpectedly. So I think tools like this could certainly facilitate those kinds of discovery. Of course, this is only the beginning, and I'm not making any specific claims, but I do believe they have the potential to help.
And how long does the MRS measurement typically take for a patient?
For a routine single-voxel examination, the acquisition itself typicallytakes on the order of five to ten minutes; total table time is longer once voxel placement, shimming, and water-reference scan are included, and editing sequences such as those used for 2-HG take longer still.
That's much quicker than most people would expect…
Yes, although it depends on the quality of the signal you want to obtain. There are always factors that can affect the acquisition, such as patient movement or the challenge of placing the voxel precisely in the region of interest. In patients with brain tumors, necrosis or edema can also complicate the measurement, so sometimes the acquisition has to be repeated. The goal is for clinicians to acquire the data in about five minutes and receive the results within another couple of minutes. That way, instead of waiting for the full diagnostic work-up, they would already have an early indication of the patient's metabolic profile and whether it is consistent with a particular type of disease - as an adjunct to, not a replacement for, the established diagnostic pathway.
How do you handle sensitive patient data in your project? Since you're working with personal health information, is it difficult to meet all the regulatory requirements for both research and future clinical use?
This is something I still need to work on. Of course, I know that people won't just upload sensitive patient data to an online platform.The goal is to develop an online toolbox that requires users to de-identify their data before uploading, and that only accepts data conforming to a defined metadata specification, so that files carrying identifiable patient information are rejected . That should make it possible to share data while protectingpatient privacy.
So far, your research has focused on neuro-oncology, but you're now extending it to neurodegenerative diseases such as Alzheimer's disease. Could this approach eventually be applied to other fields as well, for example to study the chemical changes in the brain associated with psychiatric disorders like anxiety or depression?
I'm not a specialist of psychiatric disorders, but in general, if we're able to measure metabolites in different brain states, we can compare those measurements with healthy reference values and identify potential abnormalities. So I think this is a technique that can be applied to a wide range of diseases, particularly those associated with changes in brain metabolism or neurochemistry. I know that magnetic resonance spectroscopy has been studied in Alzheimer's disease. It's also used in some studies on depression, where spectroscopy data are combined with physiological measurements or behavioral data. Together, these different types of information can provide valuable insights into prognosis and how patients respond to treatment. However, psychiatry meta-analyses have generally reported small effect sizes, suggesting that they should not be used to draw conclusions or make clinical decisions at the individual patient level. At the same time, it's important to emphasize that spectroscopy alone is usually not sufficient. To interpret the results reliably, they need to be combined with other types of data—in other words, a multimodal approach is essential.
