Pixyl: when AI deciphers medical imaging

It all began in 2008 with post-doctoral research carried out by Senan Doyle in the team now called Statify, at the Inria Grenoble Rhône-Alpes Centre, in collaboration with the Grenoble Institute of Neuroscience (GIN). The aim was to develop an algorithm capable of analysing brain imaging data and identifying abnormal or suspicious regions. “In two years, Senan had developed an effective algorithm,” says Florence Forbes, Head of the Statify team at Inria Grenoble Rhône-Alpes. “There were two options from there: stay in the lab and publish, ordevelop a turnkey tool for health professionals and patients”.

Senan Doyle, Florence Forbes and Michel Dojat, from the GIN, chose the second option, which meant creating a start-up to continue research and take the tool to the marketing stage. “This was before the AI and MedTech wave,” explains Florence Forbes. “So we had to canvass and convince people”.

Thankfully, the founders of the new start-up could count on the support of Inria, and received backing from the IT-Translation investment fund, which is supported by the Institute. This helped finance, and provided entrepreneurship training to the startup, which was initially hosted by Inria.

At the same time, the algorithm was perfected and a proof of concept (POC), a solution sound enough to attract investors and clients, was developed through continued collaboration with radiologists and neuroradiologists. “We needed both the expertise of professionals in image interpretation and the images themselves to develop our artificial intelligence,” says Senan Doyle, now CEO of the start-up Pixyl.

Developing the POC presented considerable challenges, as there was a high level of variability in the practices and methods of radiologists and the quality of the images. “We worked with some forty health structures in France, Italy, Germany, Spain and the United States to take account of this variability and obtain a robust model,” adds Senan Doyle.

With this first model established, Pixyl was finally launched in 2015. But there was still a long way to go until it could be marketed. Over the following years, the algorithm was integrated into a software program so that it could be easily used by radiologists. It also underwent further improvements. For example, the tool initially analysed 160 different brain regions. But the important information was drowned in this mass of data. “Working with radiologists and clinicians we identified the most clinically-relevant information, and now focus on twenty regions”,explains Senan Doyle.

In parallel, the team had to develop a solution to integrate the software into hospital workflows to make its use as seamless as possible and, of course, obtain the CE mark in order to market the medical device, called Pixyl.Neuro. The application was submitted in 2018 and the mark was obtained in 2019.

That year, Pixyl reached a new milestone when it won a competition organised by the French Socity of Radiology as part of the Francophone Radiology Days. “It was a competition against other companies, including major players in the sector,” explains Florence Forbes. “We were given a database of images and had two hours to process about 100 patients”. The Pixyl team were the first to submit their results and also obtained the best score.

Using technological building blocks still under development, Pixyl showed that they were able to predict the disability progression of multiple sclerosis patients in a two-year period using brain imaging. “This information could be used to tailor treatment to the patient, reducing this disability and and improving the patient's quality of life ,” says Senan Doyle. “And it has sparked a good deal of interest!” As a result, the start-up gained visibility and signed distribution and marketing partnerships with Siemens, GE Helthcare and Incepto.

Two years on, Pixyl employs 16 people and has already rolled out its solution in 20 or so health structures in Europe, despite the severe slowdown caused by the health crises. “Our brain imaging analysis product detects and categorises anomalies and compares them with images of control patients, providing the practician with information to help adapt treatment or confirm the diagnosis before the patient even leaves the consultation room,”explains Senan Doyle.“Radiologists appreciate our solution because it saves them time and provides additional quantitative information for fast, confident reporting.” The tool can be used to monitor neuroinflammatory diseases such as multiple sclerosis, migraines and epilepsy as well as neurodegenerative diseases such as Parkinson’s, Alzheimer’s, Huntington’s, etc.

To stay at the cutting edge of research, the start-up continues to work with Inria and the GIN, similar to another research project recently accepted by the French National Research Agency (ANR) for evaluating the impact of radiotherapy on certain very advanced brain cancers, as well as the AIDREAM PSPC for identifying and categorizing liver nodules.

Pixyl is continuing to grow.An initial round of fundraising of €600,000, in 2017, was followed by a second round of €2.2 M in April 2021. “This will allow us to integrate new technological building blocks for prediction into our solution, starting with the one for multiple sclerosis that we demonstrated during the competition in 2019. It should be operational by the end of 2021,”says Senan Doyle. “We will then develop similar building blocks for other pathologies according to the needs radiologists and clinicians”. 

The funding will also be used to continue roll-out in France, where the start-up hopes to see 40 healthcare facilities equipped by the end of the year. It is also setting its sights on foreign markets with a recent application for marketing approval by the FDA, the American Food and Drug Administration. From the laboratory to healthcare facilities across the world, the three founders have never been so close to their achieving their goal.