Per Fumum Endowment Fund: Could you describe the progress of your research, and what results or trends you have been able to observe?
Jean-Baptiste Masson:
“The idea was to detect traces of consciousness in patients in a coma through olfactory stimulation. We chose an unusual method of EEG analysis aimed at creating a generative model. To our great surprise, within a few weeks we obtained a first generative EEG model, using both public datasets and data from comatose patients. These generated EEGs reproduce the main characteristics of real EEGs.
We are currently working on refining these models so that physicians can no longer distinguish real EEGs from generated EEGs over short samples. Microsoft is providing us with computing resources to optimize our algorithms, and Pierre-Marie has enabled access to this infrastructure so that we can explore a wide parameter space in machine learning.
In parallel, we are developing metrics to evaluate the quality of the generated EEGs. Some are based on medical expertise — what a physician looks for in an EEG — and others rely on mathematical methods to characterize and improve generation. We have even succeeded in generating EEGs of comatose patients, whose signals are very different and often noisy due to real-world recording conditions.”
Prof. Pierre-Marie Lledo:
“For my part, I note the speed with which the synthetic EEG model was obtained. Clinical validation is currently underway, which will allow this approach to be extended to a broader spectrum of patients and to assess its usefulness for clinical decision support and for understanding brain function under olfactory and auditory stimulation. This project, innovative at both national and international levels, has no real equivalent in the scientific literature.”
Per Fumum Endowment Fund: How do patients respond to your olfactory stimulations? Are you already able to distinguish recovery trajectories?
Prof. Pierre-Marie Lledo:
“For now, EEG recordings from comatose patients are being used to validate the model. The human eye is not sufficiently trained to identify a specific olfactory signal. Previous work on event-related potentials linked to olfactory stimuli, particularly in the context of COVID, has shown how difficult it is to objectify these signals, especially in anosmic or hyposmic individuals. We therefore cannot yet distinguish precise recovery trajectories in comatose patients. The immediate objective remains the computational validation of our approach.”
Per Fumum Endowment Fund: Have you identified any odors or olfactory families that elicit stronger brain responses than others?
Prof. Pierre-Marie Lledo:
“We are currently exploring a wide range of stimuli. Extremes of valence elicit the strongest responses: vanilla or vanillin produce pleasant responses, while butyric acid (rancid odor) provokes marked unpleasant responses associated with the limbic system. Between these extremes, we use neutral odors such as lavender to broaden the sensory spectrum of the stimulations.”
Per Fumum Endowment Fund: Have there been any changes to the stimulation protocol or analytical tools since the study began?
Jean-Baptiste Masson:
“Yes, we are refining the protocol. Together with Sébastien Wagner, we have adjusted the concentration and duration of stimulation to better reproduce natural breathing. We are also introducing temporal intermittences to improve the fidelity of stimulation. On the analysis side, we are combining different mathematical domains to optimize EEG generation, while integrating medical expertise so that the machine can reproduce certain clinical evaluation criteria.”
Prof. Pierre-Marie Lledo:
“In addition, we are seeking to regenerate full EEGs for polytraumatized patients who have a limited number of electrodes. This will allow less experienced physicians to access more complete data, while preserving strong mathematical and technical interest.”
Per Fumum Endowment Fund: What will be the next steps of the research in 2026?
Jean-Baptiste Masson:
“We will generalize our approach to other hospital departments and continue developing our generative models. We will keep extracting medical heuristics from EEG analysis to integrate them into our algorithms and improve the quality of the generated data.”
Prof. Pierre-Marie Lledo:
“We also hope to explore parallel applications, for example in education, to objectify a learner’s progress. In the longer term, we could envision building a bridge between science and the arts by translating certain EEG signals into sounds or colors, but this remains a longer-term project.”
