### TLDR - Relatively simple technologies such as eye-tracking based interfaces that enable typing and/or wheelchair movement provide a benchmark for brain-machine interfaces to surpass - These eye-tracking technologies like [Toby Dynavox](https://www.tobiidynavox.com) are cheap, non-invasive and are able to achieve a record of [20-30 words per minute](https://download.mytobiidynavox.com/Literature/Software/Communicator_5/TD-C5-Brochure-011819-US.pdf) (normal typing speed with keyboard is ~40 words per minute and conversational speech is 120+ words per minute) - Non-invasive brain read/write technology doesn't have spatial or temporal resolution to power practically useful brain-machine interfaces - See [[Overview of brain input-output techniques]] to understand how EEG/MEG can be temporally precise but spatially poor while fMRI/fNIRS can be spatially relatively better but temporally poor - EEG based typing record stands at around [10-12 words per minute](https://www.pnas.org/content/112/44/E6058) and the method used ([SSVEP](https://en.wikipedia.org/wiki/Steady_state_visually_evoked_potential)) causes significant eye strain - fMRI/fNIRs detects signals that peak about 5 seconds after thought, making it practically unusable for day to day tasks (+ fMRI is bulky) - Invasive techniques such as implanting microelectrodes face a long approval process from FDA that can easily take 7-10 years and if you include R&D into this period, the total time taken from conceptualisation to launch could take 10-15 years (or more). - Even then the current record for typing using invasive techniques is [~15-18 words per minute](https://engineering.stanford.edu/magazine/article/krishna-shenoy-how-brain-computer-connections-could-end-paralysis) which isn't competitive with simple eye tracking methods - The main technical challenge with micro-electrodes is that the quality of their signals degrade over time, so long term signal acquisition from them is a challenge - Invasive stimulation techniques such as tDCS/tACs/tMS aren't popular because - They're extremely broad based (impact large areas of brain) and we don't understand their mechanisms - Their effectiveness is questionable (different studies reveal different results) - Practically, their improvement over simple techniques like talk therapy or cognitive behavior therapy isn't demonstrated - What makes progress in invasive BCI slow is: - The long gestation period (of 15+ years) from concept to being on the market - Half of that time is spent in pre-clinical R&D in labs and/or animals and half of it is spent doing clinical trials - FDA requires evidence of both safety and efficacy (over less risky-methods). It's the latter that has been difficult - Even after approval, FDA narrows the target market to those who absolutely cannot be treated any other way - Sometimes this market becomes so small that some companies like SecondSight have to shutter their launched products because available market is literally estimated to be 1500 - Clinical trials find it hard to recruit patients as right now these technologies don't promise an additional benefit over simple techniques but require an invasive brain surgery. - So patients really have to be volunteers who expect no benefit - This actually sets up a chicken and egg situation that impedes progress. - Unless there's radical benefit of a tech, patients won't agree for trials. Unless we have enough patients, we won't be able to build tech with radical benefits. - The investments into long R&D and clinical trial cycles combined with small approved target market ultimately push the price of the technology high, which makes it unaffordable to end users and puts an additional dimension of convincing insurance providers that it is worth it All this leads to something I **heard from somewhere** >"Clinical practice is 30 years behind research" Another evidence that points to challenges written above is that Facebook [abandoned](https://www.technologyreview.com/2021/07/14/1028447/facebook-brain-reading-interface-stops-funding/) (within 4 years) its much-hyped [project to develop a non-invasive brain-machine interface](https://spectrum.ieee.org/facebook-announces-typing-by-brain-project) that lets users type at the speed of 100 words per minute. In a nutshell, **existing brain-machine interfaces fare worse than relatively simple techniques** (such as eye-tracking, facial muscle detection and so on) and their chances of radical improvement are impeded by the long time and money investment into FDA approval. ### Case studies of FDA approved brain-machine interfaces #### Neuropace responsive deep brain stimulator - Invasive electrodes for treating treatment-resistant focal epileptic seizures - Started in 1997 - Clinical trial started in 2004 - FDA approval 2013 - 2019 revenue: $40mn - not profitable - IPO: $40mn - Cost per device: $50k - 2020: IPO Total time to launch from inception to commercial: **16 years.** #### SecondSight Electrical stimulation of retinal cells for people who'se photoreceptors have degraded and hence lost vision. - Started in 1998 - Clinical trial started in 2007 - Approval on humanitarian exception ground: 2013 - This happens when target market is extremely small and hence just demonstration of safety is okay - 2019 revenue: $8mn - Cost per device: $150k - 2020: shut down Total time to launch from inception to commercial: **16 years.** This technology failed because evoking precise precepts via stimulation is extremely hard, so the best this tech could make patients percieve is very crude shapes. Neurons get and send excitatory or inhibitory input to thousands of other neurons, so mapping stimulation to what is felt is not clear. Moreover, stimulation of neurons can make them adapt and stop getting stimulated. ### Case studies of brain-machine interfaces under clinical trial #### Stentrode electrodes inserted via vascular method (from blood veins) to read electric signals from the brain - Started: 2010 - Clinical trial (feasibility demonstration): 2019 - Clinical trial (for effectiveness and safety): 2021 - Expected clinical trial completion: 2024 minimum but could go up to 2026 Total time to launch from inception to commercial: **14-16 years.** #### Braingate - Started: 2000 (company started/grants given) - 2004: clinical trials began (for feasability) - 2009: braingate2 (improved technology) clinical trials began - [This study is _still_ enrolling patients (as of 2022)](https://www.clinicaltrials.gov/ct2/show/NCT00912041?term=braingate&draw=1&rank=1). They've not found all the 15 patients they need for the trial and are expected to complete the trial by 2026. - Recruiting patients in clinical trials is generally difficult but for invasive techniques like these, it's even more difficult. According to my reading, for clinical trials: - Nearly half of all clinical research sites are under-enrolled - 11% of sites fail to recruit even a single person - 80% of clinical trials are delayed or closed Total time to launch from inception to commercial: **25+ years.** <iframe class="signup-iframe" src="https://invertedpassion.com/signup-collector" title="Signup collector"></iframe>