TheraPanacea: Mathematics in the Service of Radiation Therapy
From research in mathematics and artificial intelligence to clinical application: TheraPanacea provides a concrete example of what can be achieved through collaboration between engineers, researchers, and healthcare professionals. Founded in 2017 as a result of research conducted at CentraleSupélec and Inria Paris-Saclay, TheraPanacea develops solutions to improve the planning and adaptation of radiation therapy treatments. In this interview, Nikos Paragios, co-founder and CEO of TheraPanacea and professor at CentraleSupélec, discusses the project’s origins, the work carried out with healthcare professionals, and the challenges involved in transitioning from research to technology deployed in healthcare facilities.
How did the idea come about to apply your research to the medical field and then co-found TheraPanacea in 2017?
Image
I’ve always been convinced that basic research only reaches its full potential when it tackles real-world problems. At CentraleSupélec, within the Digital Vision Center, and with the GALEN team at Inria Paris-Saclay, we spent nearly fifteen years developing mathematical methods for analyzing medical images.
In the mid-2010s, after years of fruitful academic research, a question arose: why were these results confined to academic publications? With the support of Paris-Saclay University, we evaluated several areas of application—diagnostics, radiation therapy, and robotic surgery. Radiation therapy emerged as the clear choice, both because of the maturity of the need and because it aligned with our expertise. My personal experience—several members of my family have undergone radiation therapy—rooted this decision in something deeper than a strategic calculation. TheraPanacea was founded in 2017 out of this convergence of scientific capability and clinical urgency.
What medical need were you initially seeking to address, and how do your technologies specifically contribute to patient care?
Radiation therapy uses high-energy radiation to destroy tumor cells while sparing healthy tissue. Preparing each treatment requires delineating the tumor and each at-risk organ on the patient’s images, then designing an optimal radiation plan. This work is still largely manual. It takes hours. And the quality varies considerably from one practitioner to another, and from one center to another.
Our platform, ART-Plan, automates this entire process: anatomical segmentation, treatment plan generation, and daily adaptation to changes in the patient’s anatomy. The time savings are significant. But the key benefit lies elsewhere: by embedding best clinical practices into our algorithms, we guarantee a consistent level of quality. A patient treated at a regional hospital receives the same level of care as a patient at a major university medical center. It’s a matter of equity as much as performance.
How have you worked with doctors and healthcare institutions to develop and refine your solutions?
I often use a metaphor. Writing an excellent scientific article is like modeling the tip of the iceberg. Building a product is about understanding the submerged part. A successful product is one that masters the 80% that no one sees. In practice, this rests on three pillars. The first: identifying the real clinical need—not what’s scientifically elegant, but what makes a difference in the day-to-day work of a radiation oncologist or medical physicist. The second: accessing data that reflects the true complexity—the diversity of diseases, anatomies, and protocols. This is where partnerships with hospitals are crucial. The third: integrating the solution into the existing workflow, seamlessly and without disruption. Ironically, of these three challenges, technology is the easiest to solve.
What changes when you move from a research innovation to a technology deployed on a large scale? What challenges did you have to overcome?
Scaling up changes the very nature of the problem. In research, we optimize a metric on a controlled dataset. In production, we guarantee a result for every patient, in every center, regardless of equipment, protocols, and local practices. Clinical variability is immense—it is the real adversary, not algorithmic complexity. First challenge: the regulatory framework. A research algorithm is not a medical device. CE and FDA certifications, quality systems, traceability, post-market surveillance—it’s a profession in and of itself. Second challenge: integration. Every center has its own IT ecosystem, equipment, and workflows. An effective model that disrupts the workflow will be rejected. It must be made invisible within the clinical routine. Third challenge: trust. A doctor will only adopt a tool if they can verify and control it.
We do not replace the clinician. We give them a tool that makes them faster and more consistent—the final decision rests with them.
What does this journey reveal about the role that CentraleSupélec engineers and researchers can play in transforming the medical field?
TheraPanacea is a prime example: a mathematician from CentraleSupélec can directly contribute to improving the care of cancer patients. The medicine of tomorrow is computational—modeling, optimization, and large-scale data processing. These are precisely the skills the school develops. What is often lacking is not technical ability—it’s exposure to clinical problems. My message to students and researchers is simple: go to hospitals, talk to doctors, and observe the challenges they face. The most important problems don’t lie in the complexity of the mathematical solution—they’re right there at the patient’s bedside.
Learn more:
- Therapanacea
- Above, an illustration of adaptive radiation therapy: a comparison of the initial treatment plan with the plan adapted to the patient’s anatomical changes.
