We’re happy to announce that BrainTrip has been chosen as the best AI-Driven Solution by the Abarca Prize.
Our approach to AI isn’t that complex or fanciful, but it is smart. It’s smart because we carefully prepare the input data. We do a lot of clever data pre-processing to improve the signal/noise ratio, and we take into account confounding variables. We need to do this, otherwise we wouldn’t know what we’re actually measuring. In our example, the BrainTrip Dementia Index (BDI) could easily be measuring age instead of neurodegenerative cognitive decline – older people are more likely to suffer from dementia. But the BDI doesn’t measure age, because we correct for age and several other variables before the AI stage. We use a relatively simple neural network with just a few layers. This is by design. The more complex and multi-layered the neural network, the more data hungry it becomes. When there’s data in abundance, this isn’t a problem. Cat recognition software is readily available because people are more than happy to upload photos of their feline friends. They even label these photos with hashtags. It’s like Instagram was made to train AI models. The same can’t be said for brain data. Sure, there are some databases, but none of them were good enough to develop a tool for early dementia detection. We had to collect our own data set – about 2000 BDI recordings. We have one of the biggest EEG-based datasets in the world, but when compared to the number of cat photos in existence, it’s tiny. If we took the same approach as cat-recognition AI to make the BDI work with 95% accuracy, as it currently does, we’d need to do an EEG recording on every human on the planet. Why? Because of data complexity. The more complex the data, the more of it you need. And, as you might imagine, brain-wave data is quite complex. (We even did the math to figure this out. And by we, I mean Jurij Dreo, I’m not smart enough to do math). Because of this inherent limitation, we had to figure out a different approach. We had to make the BDI work with hundreds, not millions of data points. Which we did! Thanks to Jurij Dreo and our product development team, we’ve developed a system that takes in raw EEG recordings and spits out a score of the brain’s cognitive health. We still use AI, but it isn’t an impenetrable black box. It’s just another tool. One of the many data-processing steps needed to produce a highly accurate result. Thank you Abarca Prize for recognizing us and awarding our smart approach to AI. Thank you Ventures Thrive for supporting us over the last months. Thank you EIT Health InnoStars for your ongoing support with the #HICEE program. And last but not least, thank you Hello Tomorrow for choosing BrainTrip as one of your Deep Tech Pioneers. https://lnkd.in/dC9kM63r