Problem definition: Brain computer interfaces are well underway to becoming the healthcare innovations of this generation and the next. The development of these devices provides opportunities to restore function in conditions previously thought untreatable, such as motor neurone disease. The creation of a purpose-built BCI blockchain at the preclinical stages will allow for optimisation and implementation of best practices from patient zero, in the fields of data security, neural sovereignty, informed consent, sustainability and functionality.
As investigators and interventionalists, we are bound by two constraints: the limit of existing knowledge and capabilities, and our own hypotheses for the future. With the passing of each age of neural investigation, the fraternity bestows their advancements onto the next generation. Initially, these advances were primarily anatomical – through Brodmann’s classifications or Cajal’s silver drawings[1]. Clinical investigators built upon that knowledge, contributing descriptions of regional functionality. Multidisciplinary STEM collaboration led to the research continuing today, such as O’Keefe’s measurement of the hippocampal geolocation[3]. This crescendo of knowledge has now led us to the precipice of restoring and enhancing neurological functions once thought lost.
Recently, Oxley et al described his success implanting the first in-human motor prosthesis, with promising functional return[4]. Advances such as these signal the next step in clinical neurointervention - brain-computer interfaces (BCIs). These devices vary in nature and complexity, from deep brain stimulation (DBI) which augment the activity of a certain region of the brain for functional response, to machine learning-facilitated neuromodulatory devices which are able to predict when misfiring is occurring, and remedy that misfiring, such as that described in DARPA-funded research to enhance memory[2], and Oxley’s motor prosthesis.
So, what is it that we must consider at this juncture? The predominating concerns and limitations of these therapies include basic considerations such as safety and efficacy, but also entail wider ethical and technical issues such as security, data storage and signal modification, and eventual scaling of such devices. A number of people have already considered blockchain as a natural response to these concerns, but as of yet, there is limited research on how these models may be structured in future.
Currently, the alternative to blockchain technology is either not storing this data for future use or providing external sources complete access and ownership of patient’s neural data on company or institution-owned centralised servers, which is open to exploitation and sale, and subject to server costs, regulation and interference. Blockchain technology may allow patients to reduce vulnerabilities to therapeutic companies and increase patient trust through enhanced security and autonomy.
Principles of a BCI Blockchain
The oft mentioned ‘blockchain trilemma’ speaks of three principles that comprise a successful blockchain. These principles include decentralisation, security and scalability. These principles are also compatible with a competent model of storage and computation for brain computer interfaces. The term ‘blockchain trilemma’ speaks to the difficulty of combining all three principles in one blockchain, however purpose-built blockchain architecture can allow for prioritisation of the principles relevant to any particular project.
Decentralisation
Decentralisation, or distributed ledger technology, distributes transactions or interactions over a number of nodes, meaning no central party, or company, would have exclusive, complete control over any patient’s data, rather those with access to the blockchain would all be able to access proof of transaction, which may contain a small temporal bin of relevant deidentified data. By using a private* blockchain to store temporal bins of anonymised brain activity, patients are able to retain ownership of their data, while still allowing research and reporting capabilities on their implants. It also means that implant specifications can be governed by smart contracts, which cannot be altered without patient permission, allowing dynamic patient consent.
Security
There is some mitigation of risk by utilising a closed-system, purpose built blockchain such as this one. This may be achieved by limiting access to source code to contributing research institutions or hospitals, distributing tasks across multiple layers, building smart contracts to specification, employing the use of trusted neuromorphic chips, and ensuring patients retain their own private keys (or via proxy through their treating doctor). Building a purpose-built blockchain not only allows for increased security, it also means you are not limited by another blockchain’s traffic, which can slow processing times and speeds, and increase cost. It also removes the possibility of compounded risk through unrelated exploits on an external blockchain, or risk associated with a developing team not associated with the project.
Scalability
Scalability of this blockchain is achieved by offloading complex computation off-chain on neuromorphic chips, and on layer two solutions. As this blockchain scales, it should actually improve the treatment for patients, as every input into the blockchain allows the machine learning algorithm to learn, which may impact development of neuromorphic chips in future, as well as improve reporting on their accuracy at a patient level.
Along with these three principles, computational speed can be addressed using a division of tasks through error recognition at the neuromorphic chip, smart contract technology verification, and layer one computation, learning and reporting. Learning may also be offloaded onto layer two.
This Blockchain Model: The Nitty Gritty
Neural information is to be collected through a neural array, and communicated to a neuromorphic chip, where basic neural computation, identification of pathological processes (error) will occur, and stimulation will be generated. At the point of potential error, a snapshot of neural information at that time point and a temporal bin each side of the error will be collected and transmitted through to the blockchain (layer 2). A smart contract will then verify or deny that it is a pathological process, based on integrated data such as length of error, amplitude etc. This may either allow verification of BCI intervention to the pathological process, or purely as a reporting tool. In the case of using the blockchain as a verification tool, the chip will receive input from the layer 2 to proceed with intervention in the form of a targeted electrical stimulus to the region of interest.
This snapshot of neural activity (whether verified or denied) will then be communicated to level one of the blockchain for storage, computation and export via numerical string based on electrode input location and input strength within that time period. In times of healthy brain firing, defined as no reported error, the chip will remain in ‘monitoring’ mode and no information will be transmitted to the blockchain.
This model has been designed to be modifiable for multiple applications of BCI as treatment for disorders of both excitatory and inhibitory control. Some applications are seizure recognition and intervention, tremor suppression, neurodegenerative monitoring, with maintenance of progressively degenerating synapses, and future applications such as memory and focus optimisation without the use of systemic pharmaceuticals.
Figure One: Figure demonstrating BCI:blockchain model, illustrating inputs between neural array, neuromorphic chip, and layers of the blockchain. ML may also take place on layer two.
Base Layer: Layer One
Layer one of this blockchain model has three functions: storage (for research and machine learning dataset) and export.
This layer, firstly, stores all neural activity with suspected error – with a temporal bin on either side. This data will contain both verified (confirmed error) and dismissed (rejection of error calculation) neural activity, essentially acting as a store of all data that returned a positive (error) response at the level of the neuromorphic chip. This data forms the input to a simple machine learning pattern-recognition algorithm. This algorithm will be able to identify features that the neural data correctly, and incorrectly labelled as error. This information may then be exported off-chain for research and development purposes – particularly to refine the parameters of the neuromorphic chip. Depending on the capability of the chip used, this may ideally be used to modify chips already implanted but may also be used for future chip generations. In order to feed back to the chip, this Layer One must have built in capability to communicate to Layer Two, and onward to the chip.
Verification Layer: Layer Two
The purpose of Layer Two is to receive input from the neuromorphic chip, and depending on chip capabilities, identify “error” from information received through the neural array and neuromorphic chip, or alternatively (in the case of a more sophisticated chip) verification of the “provisional error” suggested by the chip.
In order to achieve this, the Layer Two must be able to receive information from the neuromorphic chip, perform a verification task, communicate response back to the neuromorphic chip, and communicate the recorded “provisional error” and response through to Layer One.
Additionally, the Layer Two must be modifiable. In development and maturation of the blockchain, this layer will be modified for three purposes, reducing to two purposes as keys are discarded in order to achieve improved security as the blockchain reaches a sustainable size and the algorithm is sufficiently tested. The first reason for Layer Two modification is for clinician-advised modification, in order to increase or reduce intervention at the request of the patient. This is a crucial factor, in order for patients to retain the power to remove consent for treatment, or for other patient-specific reasons for intervention cessation or modification such as potential adverse reactions. The second reason for Layer Two modification is for developer-advised modification, such as improved functionality through updates, patching, or fixes. The third reason for Layer Three modification is for pattern recognition algorithm-specific modification, in case of automated response from layer 1 learning through to intervention via neural array.
Communication between structural layers
Dynamic smart contracts between the chip and the layer 2 allow for modifiable input, from the everchanging neural input and any required modifications by the treatment team based on layer 1 learning and patient experience. These modifications would allow for patient-specific increased or decreased trust in chip computation, allowing for more conservative treatment while the chip is still primitive, resulting in ability to reject intervention at a higher rate when signals are not as clear, or reduce the power or frequency of intervention. As these intervention requests are recorded on the layer 1 blockchain and certainty of conditions to intervene is increased over time, chip trust is increased from layer 2, resulting in stronger responses as well as more accurate rejection or approval of intervention.
Discrete units on layer two allow for one patient per unit, to allow patient-specific trust modification of chip responses. These discrete units communicate to layer 1 to allow for spatial, temporal, frequency and amplitude-based optimisation while maintaining dynamic modification at layer 2. The discrete layer 2 unit per patient model also allows for patient-specific layer 2 script updates, allowing for increased patient control and autonomy, only allowing script updates with patient informed consent. In future, as arrays modify and improve, this information may be used to optimise placement surgery, by modifying electrode choice and location.
Future Directions
The elegance of a chip-optimising blockchain BCI is that the blockchain’s layer one will over time contain the data pool that will inform and train the machine learning algorithm that will improve the implanted neuromorphic chip. This may mean that the longer a chip is implanted, and as more chips are implanted, the smarter the chips will become. This progressively and increasingly solves interventional inaccuracies.
The provisional decision to generate an intervention could be decided at the chip level or at the layer 2 level, as long as confirmation occurs at layer 2. Eventually in future generations of this technology, accuracy may even be improved by allowing learning intra-personally through the chip and layer 2, along with interpersonal learning.
In future, this model may provide an avenue for diagnostics of neural pathology, as well as treatment, as the algorithm will allow for the generalised pattern recognition of impaired firing. At this time, it may be able to recognise error, as well as replicate a standard expected firing pattern.
Conclusion
It is our responsibility as the investigators and medical professionals of today to ensure that our new technologies and therapies are safe at time of implementation, with risk minimisation strategies in place. The blockchain model suggested provides a way for patients requiring neuro-interventional treatment to maintain autonomy, and power over their neural data. With a wide range of biocomputational tools currently in development, the decentralisation of the resulting data is an important step towards data security, patient safety and personal autonomy.
References
1. Cajal S (1903) Un sencillo método de coloración selectiva del retículo protoplasmático y sus efectos en diversos órganos nerviosos. Trab Lab Invest Biol 2:129–221
2. Hampson RE, Song D, Robinson BS, et al (2018) Developing a hippocampal neural prosthetic to facilitate human memory encoding and recall. J Neural Eng. doi: 10.1088/1741-2552/aaaed7
3. Morris RGM, Garrud P, Rawlins JNP, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297(5868):681–683
4. Oxley TJ, Yoo PE, Rind GS, et al (2021) Motor neuroprosthesis implanted with neurointerventional surgery improves capacity for activities of daily living tasks in severe paralysis: First in-human experience. J Neurointerv Surg 13(2):102–108
How completely and totally brainy are you Stephanie Bazley. Beyond. Beyond!