Synthesizing fMRI scans using EGG data to detect Alzheimer’s and other brain diseases
Last January me and my team for the Global TKS focus hackathon, imagined a solution that will allow the detection of Alzheimer’s and other neurogenerative diseases and tumors that normally require an fMRI with a simple EEG cap.
Here is an article explaining what we’ve done and how we’ve done it.
fMRI cost often more than 400,000€ and around 600€ per hour, this is extremely expensive but needed given how fMRIs are useful to detect a huge variety of issues and diseases. But unfortunately, lots of developing countries cannot afford to have fMRI or given their limited number cannot scan every person with symptoms.
That’s why we imagined an AI that will be from a simple EGG live recording to determine if a person has Alzheimer’s!!
Early detection of those diseases can make an important impact on the treatment and people’s life. Our solution would allow detecting every person that would be at risk in 10 minutes with a 300€ EEG cap instead of a 400,000€ fMRI scan.
Now the big question that you’re asking yourself, is how can we generate artificial fMRI scans with only simple EGG data?
As you can see above EEG and fMRI scans are very different, actually, EEG has a very high temporal resolution and low spatial resolution, and fMRI has a high spatial resolution but low temporal resolution.
This is where the idea of using simultaneously recorded EEG and FRMI data sparked in our heads. In fact, both of those technologies are overlapping because they are scanning the same object., a brain 🧠
So we got the idea of using a GAN (generative adversarial network) to recreate the fMRI scan with the EEG data.
A GAN is a set of two artificial neural networks competing in a zero-sum game to win. We more specifically used a WGAN, to improve the quality and accuracy.
The process:
We haven’t actually got the time to build this as we got 24hours, this is only what we think would work
We’re first doing some little modification to our data to allow our WGAN to work well, we’re performing an STFT (Short-time Fourier transform) on the EEG data and downsampling a bit the MRI data.
As our data doesn’t have the same temporal resolution, we need to “sync” them with a process that we’re calling BOLD shift emulation.
After that data are split again and sent to the next part.
This is the diagram of how our WGAN is working, I won’t go into details to keep that short but here is a brief explanation:
Our GAN is composed of two main components: a discriminator and a generator.
The goal of the Discriminator is to recognize instances synthesized by the Generator, if it is able to do so then a penalization is given to the Generator. On the other hand, if the Discriminator does not recognize those synthesized instances, a penalization is given to Discriminatorn itself.
Here pr discriminator and generator are materialized by our two lost functions the L_c one and the L_r
With this process, we’re able to have an artificially synthesized fMRI scan from our EGG data
This part is our other AI that is doing the diagnosis, in this case, we’re detecting Alzheimer’s from our synthetized fMRI scans.
To do so, we’re using a CNN (convolutional neural network). This is a classic architecture used for images. This CNN would have been trained on detecting Aizhamenr on real fMRI scans downsampled to the same level as our syndicated one. (this has already been done serval times, see Sarraf and Tofighi or Al-Khuzaie et all.)
The future?
Our solution could have a huge impact on the life of millions of people across the world and save millions of dollars for the health system of various countries. There is only one little condition to that… it needs it works!!!
My team and I are pretty confident that our solution could work with more data. The currently available datasets are really small and in low resolution, we think that more datasets with higher quality and more samples could allow this kind of solution to exist.
And if you’re asking: wait wait Jules are you going to develop that idea?
I might I might… If you have any ideas or if you would be interesting to work on this project, message me!📨
I’m Jules Padova, a 16yo TKS innovator who is trying to improve our world through AI 🤖, neuroscience 🧠& biotechnologies 🧬
If you want to follow my projects, follow me on Linkedin, Twitter and subscribe to my newsletter!
