Calculation

So, if I give 6 bits to encode letters, such as 26 English letters or 28 Arabic characters, and the rest of the available 6 bits for punctuation, numbers..., Now icEEG is often sampled nowadays at 4096 Hz (which means samples per second from each channel) or higher in advanced centers, typically from a different number of electrode contacts, but for this example, let's say 250 electrode contacts or equivalent. Each electrode contact represents activity from approximately 1 cm square from the brain's surface and tens of millions of neurons, so we have a connection from 250*4,096 = 1,024,000 points/sec of information from approx 250 cm^2 area of the surface brain. On a side note, the typical brain surface is 1,500 to 2,000 centimeters square. But that is not the only story: This number is multiplied by 16 bits, which is the dynamic range that will enable us to decode and store with fidelity the voltage at each given time point = 1,024,000 points/sec * 16 = 16,384,000 bit/sec. This is when things may begin to get a little fascinating: let's divide this number by the 6 bits necessary to generate any text in Arabic or English, for example:

We have 16,384,000/6= 2,730,666.67 etext/sec of data. Let's divide that by the average number of letters per word: 5.75 = 474,898.551 individual words. This is where things start to get a little bit even more enjoyable. A typical ebook with 10K-100K words, let’s stick with a very conservative 11K in this instance, which implies that the EEG readings were taken from only 10-15% of the brain's surface, produces the equivalent of 43 ebooks per second of abstract voltage data... amazing.

On average, icEEG produces the equivalent of a 43-unique full-featured EEG book.

Do you know that

The typical duration of icEEG evaluation is 2-3 weeks. how many books is that
While attending to the standard of care and following guidelines to the best possible, clinicians screen 24 hours' worth of data in under an hour.
In the computational age/era of digitization and data, we may detect beauties, links, information about brain dynamics in function, and disease that would have been impossible otherwise. Not even an intelligent human eye can start to decipher such complexity.

This is the amount encoded from free-running recordings of abstract voltage data in time series. Imagine possibilities when factoring in spectral content (e.g., high-frequency neuronal/basket cell firing, infraslow, conventional frequencies, theta cognitive, alpha/beta: functional. Possibilities include factoring in tasks time of the day, multimodal incorporation such as the chemistry of neurotransmitters, single-unit recording, blood flow coupling ... etc.

Implications and Applications

The rumor has it that at the American Clinical Neurophysiology's presidential address decades ago, the speaker believed that with EEG, humanity had explored everything feasible, and it's time to move on to another technology. Note EEG was successfully recorded for the first time in the 20s of the past century. It only took a few more years and EEG digitization, eliminating the physical-mechanical limitations of analog EEG and opening the way to monitoring neuronal activity in addition to the more prolonged synaptic potentials and a whole new suite of spectral analyses using over the shelf clinical and household equipment. Nowadays, the most remarkable breakthroughs on earth are in neuroscience and brain-computer interface, outer space, and digital discoveries are still fascinating!

In recent years, and with the digitization of EEG, neuroscience has witnessed a resurgence of multivariate analysis techniques. Deep neural network (DNN) models have been successfully applied to neuroimaging data and interpreted as if they were 'brain-reading.' There have been impressive efforts for Language, Image, and EEG Decoding Using Deep Neural Networks with Enhanced Explicit/Implicit Salience Matching for Clinical Brain-Computer Interface Applications and direct EEG recordings and decoding at the letter the pixel level from direct brain data. How AI and big data in EEG can help clinical practice: The endless possibilities have been applied practically in helping the disabled by enabling direct brain-computer interactions bypassing paralyzed body parts. e.g., recent news of a full tweet via a brain chip, prothetic control in quadriplegic, and many other efforts. Remarkably advances in the private sector were reported, such as in Neuralink. This, without a doubt, will have an impact on the future of the human-machine hybrid and when it will happen. Is the grass greener somewhere else?

Our Recent Effort on AI-based processing of EEG signal

Did this altogether convince you about the endless possibilities with icEEG evaluation in patient care and research? How do you feel? In the following paper, we compared an AI-based: The popular SVM, and non-AI-expert-designed computational methods. The central sulcus is one of the first consistent structures described in the human brain, credit to Luigi Rolando 1817 A.D. (little kept secret, he concluded that the primary motor centers in the brain were in the cerebellum). Its localization central to any surgical practice as it marks the crucial sensorimotor functions, e.g., homunculi, that must be preserved in neurosurgical planning, e.g., tumoral or epilepsy surgery. Much has been accomplished in mapping it, whether by direct cortical stimulation, evoked potentials, functional imaging, among other techniques. As proof of the concept, we show in this study that it is possible to decode a significant amount of information from a 6-minute free-running EEG during sleep without participation from individuals/tasks or stimulation. This has not been done before. The application of the method does not require particular expertise. e.g., To find, for instance, where is the hand area. Typically, it may take hours to accomplish this task per standard of care. This is fantastic and opens the doors for endless possibilities. Not only that, we showed an informed expert design of the computational metric performed on the same footing as modern machine learning. I truly hope the future will bring more interest in the area. Isolated efforts in this age and era are like an expanding bubble in one's head with a predictable outcome.

So next time someone claims they figured EEG out, challenge them to think harder.