Meet the Expert: Dr. Tobias Lindig

Founder AIRAmed

Neurologist, Neuroradiologist, University Researcher

Both a neurologist and neuroradiologist, Dr. Lindig founded the German medical technology company AIRAmed. The start-up recently announced FDA 510(k) clearance of AIRAscore, a medical image management and processing system providing relevant brain volumetry data to assist physicians in early detection of Alzheimer’s and other dementias. The first person to test the software on himself, we talked with him about the impetus behind this AI-driven software and its future potential.

What is your personal story about choosing medicine and founding AIRAmed?

From an early age, I’ve been fascinated by the brain and how it works. After completing my studies, I initially pursued a career as a neurologist. However, my research projects consistently were at the intersection of neuroscience and MRI imaging. To delve even deeper into this special field of MR imaging of the brain, I decided to expand my expertise and became a radiologist, specializing later in neuroradiology.

The fields of neurology and radiology differ significantly: In neurology, we take an objective approach. We measure and quantify nearly every aspect. Objective data such as test results and numerical values play a central role in diagnosis and treatment. In radiology, we have very few objective measurements. Instead, it’s all about the visual analysis of images; diagnostics are dependent on the expertise of the diagnosing physician.

Until recently, precise measurements from 3D image data were technically challenging.
That has now changed. Thanks to technological advancements, particularly AI, we are able to reliably measure MR image data within minutes. At AIRAmed, we leverage these capabilities to detect reductions in brain volume. This plays a crucial role in early detection and differential diagnosis of Alzheimer’s and other dementias.

By analyzing these measurements, we can identify whether a decrease in brain volume is age appropriate or if it signifies the onset of a disease. Furthermore, these measurements enable tracking the effectiveness of therapies throughout the course of the disease. Coming from academic research, we founded AIRAmed to bring this added value into clinical practice and to where it is mostly needed – to patients and their families.

Does radiology accept the measurements calculated by AI or is there pushback?

Pushback isn’t the right word. For most radiologists, there is no question that AI surpasses the human eye in many aspects. Using brain volumetry as an example we can provide precise measurements that wouldn´t otherwise be available as diagnostic parameters.

Within the radiological community, there’s definitely a growing consensus that AI will soon be integral to radiological interpretation. As the famous quote goes, ‘AI won’t replace radiologists, but it will replace those who don’t embrace AI support.’ However, we’re still in the early stages, and it’s natural for some to remain skeptical. Technical challenges persist, clinical validations take time, not every AI software on the market is good software and most AI solutions are still first-generation products, including our software, AIRAscore. At AIRAmed, we’re confident in our path and expertise. However, it will likely take some time until even the most skeptical colleagues are convinced that high-quality radiological interpretation without AI will soon become nearly impossible in many cases.

How would you frame the problem that your technology is solving?

The brain volume of every individual decreases with age. However, it’s nearly impossible to determine with the naked eye whether this decline is typical for aging or if an additional pathological process is at play. Our software, powered by AI and neural networks, can segment relevant brain structures and tissue areas, allowing us to precisely calculate the volume of each brain region. By comparing these findings to a healthy cohort of people of the same age and gender, we can differentiate between normal aging and abnormal changes. If a disease is present, we can diagnose the specific neurodegenerative condition and monitor treatment success over time.

Are you planning to use your software in clinical trials?

Yes, of course. We are already using AIRAscore in clinical trials with different research partners and CROs.

What was the input to your AI? Healthy brain scans? Diseased brain scans?

Our AI system is trained using both healthy and diseased brain scans across a wide age range (from 6 to 100 years). We incorporate data from various MRI scanners and sequences to train our neural networks. During this training, the algorithm learns to accurately segment different brain regions and tissues, enabling precise volume measurements. Next, we establish a reliable reference cohort from the population being examined—such as Europe, North America, or China. Each region requires separate reference datasets. To create a robust reference cohort, we collect thousands of brain scans. The quality of the reference cohort is as equal to the quality of the total product as the AI algorithm itself.

Where does the training data come from?

We compile data from various sources, both internal and external. Our clinical research partners provide pseudonymized clinical data from studies. Additionally, we curate and refine data from large open-access cohort collections.

For your FDA clearance, was that algorithm for North American images?

Yes. To get the FDA clearance it was necessary to use North American patient brain scans.

Is it by race or regions that brains differ?

It’s not really clear right now. The tissue is always the same independent of race. We all have the same brain to start with, however for example depending on our language, the brain can be smaller or larger. Because the Chinese language is more difficult than European languages, that makes a difference in the respective brain region which affects volume and needs to be considered. These differences are small and not highly relevant; however, to achieve the greatest possible precision, the correct reference collective should be determined.

How long did it take to gather and sort out this data?

When we started in 2019, it took about a year to get our first algorithms working; now it’s ongoing and we are continuously improving the product. As far as the product is concerned, as soon as we finish developing a new version of the product, we show that it’s working and that it’s safe. Each time we release the latest product version to our customers, they automatically get the newest updates. Then we’re in the background working on the next release.

What are the requirements for updates?

For major releases, we submit an adjusted 510(k) to the FDA for approval, but not for bug fixes or minor changes.

What will it mean when you are successful?

It will mean that we are the most reliable product on the market, and that all patients worldwide have access to our tools. For me personally, it means that everybody becomes aware of the technique and the benefits of quantifying brain MRIs – effectively bringing brain volumetry into routine screening. In addition to brain volumetry, in an ideal future, at the age of 45 or 50 everyone gets a baseline MRI, and every two to five years thereafter a follow-up screening check is done for structural changes to avoid diagnostic and treatment delays.

What do you see as the future for diagnosing and treating the brain for dementias?

If a person is fading away, you can’t really bring them back. With early detection, much can be done to delay or prevent disease – medications, care with lifestyle, and most importantly, it’s known that something has to change, or the future is limited. Time is of the essence to do something in the next two to three years, not in 10 years. While not everyone may want to know this information, for those who do, we are able to give them that information with good certainty of whether there is something of concern or not.

Finally, what are your thoughts about the brain atlas project – the ten-year NIH research project that cost $500 million?

The new brain atlas is exciting work. Unveiling commonalities of brain structures across species – from mouse to man – is essential for a deeper understanding of similarities in the complex space of the brain. That complexity is the biggest challenge in gaining this deeper understanding of how the brain works and how certain cell types are mapped in the brain architecture. However, as there are more connections between brain cells in the human brain than grains of sand on earth, understanding brain connectivity is the real challenge and is yet unsolved.