Five of the 16 doctoral students competing for the Leon Labor Exchange on May 31, 2022 were awarded by jury, public and high school classes during the French final of the popular competition “My Thesis in 180 Seconds”. Presentation of their research topic in three minutes. Alfani Middlett, a doctoral student at the University of Grenoble Alps, won first prize for his presentation “Treatment of Obstructive Sleep Apnea by Continuous Positive Stress”, by Manuel Tunon de Lara, President and President of France Universities, and Antoine Petit, President and CEO. CNRS. He will defend France in the international final of the competition scheduled for October 6, 2022 in Montreal.
Sciences at Avenue: What training did you follow and what led you to work on sleep apnea?
Alfani Middlelet : I did an engineering school at the University of Technology Compiègne, with specializations in IT, data modeling and decision support. Then I did an eight-month internship as a “data-scientist” at Probes, a company in the Grenoble region that develops Taylor-built artificial intelligence solutions. After that, I am working on a thesis at the HP2 Laboratory (Hypoxia and Cardiac and Respiratory Physiopathologies) with the Industrial Chair “My Way to Health” at MIAI’s Multidisciplinary Institute of Artificial Intelligence in Grenoble. I wanted to deepen my knowledge, dig into modeling methods and apply my subject to a field that is closer to my heart in terms of health. Sleep apnea is a major health problem and sleep has always fascinated me. In addition, the possibility of interacting with doctors was interesting.
What is this persistent positive airway pressure sleep apnea treatment and how does it work?
It is a nasal mask that is worn while the patient is sleeping and is attached to a machine that sends compressed air to keep the upper airways open. Obstructive apnea occurs when the pharyngeal soft tissues become loose and, depending on the degree of obesity, body position during sleep, complete or incomplete collapse of the neck may occur, with reduced or obstructed airflow.
For my work, I use the Apnea + Hypopnea indicator, which makes it possible to assess the level of control intensity of Obstructive Sleep Apnea Hypopnea Syndrome despite the use of this treatment. Hypopnea is a decrease in airflow, without complete obstruction, but which still causes hypoxia, a decrease in the level of oxygen in the blood. This residual apneas and hypopneas can be detected using an algorithm applied to the machine. Then I get daily data of each patient while observing a few thousand patients. We try to analyze the different trajectories of treatment response, the effects of underlying pathologies and the effects caused by changes in the mask or changes in stress levels in disease control.
While such data has only been collected since 2018, the scientific literature is still quite poor and the existing articles consider the average data in one or six months. The challenge with my thesis is to model daily data as a time series. Data monitoring remains complex because several tools have been developed to monitor residual apnea during sleep. Typically, doctors visit their patient once a year and sometimes realize too late that apnea persists despite ventilation support, which can lead to early abandonment of treatment or cardiovascular complications.
You mentioned the risk of heart problems parallel to sleep apnea. How is it identified and can it be identified?
We are trying to show that this data of the connected object is a resource, because it is a snapshot of the patient’s condition on a daily basis. If he frequently develops residual apnea and / or hypopnea, it may be a sign of an event cardiovascular disorder and / or the emergence of central apnea: it is no longer a mechanical obstruction, but a nervous disorder respiratory drive. We can detect this with the help of this machine, which is not the case with other tools. I also study documented raw airflow, for example, in patients with chain-stokes breathing, such as an abnormal breathing rhythm that changes the duration of apnea and hyperpnea (increased respiratory amplitude). We show that among these people, the length of cycles and their characteristics are more specific to heart failure than other conditions. If we applied this signal processing to the current machine, we would be able to detect the presence of a cardiac event early on. It is common to be diagnosed with cardiac disorder after an acute event like a heart attack, so it is important to be able to detect this type of pathology as soon as possible.
Artificial intelligence makes it possible to realize that the patient needs to be considered as a whole. In my laboratory, we maintain strong collaboration between disciplines: I work with pulmonologists, sleep specialists and physiologists. They talk to me about the problem, give me clinical knowledge. Without it, I wouldn’t understand the data. In addition, the laboratory includes an e-health chair and an artificial intelligence chair at the MIAI Institute, which aims to apply innovative analysis methods to this type of health data.
Why do you want to participate in “My Thesis in 180 Seconds”?
I wanted to be clear in the presentation of my work. Speaking in front of an audience was also a personal challenge. What inspired me was the training in “My Thesis in 180 Seconds” that we received from Grenoble actor Ludovic Lecardia. I’m still surprised to win this edition, I still think it’s hard to go one step further and know what made the jury and the public happy. I know sleep apnea can talk to people, but it’s always hard to show that we’re working on it.
What do you want to do at the end of your thesis?
Generally, I defend my thesis in November 2022. I’m not sure what would happen next, I really liked teaching, so why not continue this way. But I would like to continue my expertise in data analysis services for health, it is in full expansion and there is much more to do. I think it’s a highly rewarding environment for evolution, with many initiatives emerging and ending when doctors were reluctant to face artificial intelligence. However, it should be noted that these issues also depend on the involvement of manufacturers, home service providers and machine manufacturers in the implementation of these data monitoring solutions.