In the realm of the hybrid brain Human Brain and AI Hoda Fares1 Margherita Ronchini1 Milad Zamani1 Hooman Farkhani1 Mich ela Chia ppalone2 Emre Neft ci3 and Farshad Moradi1

2025-05-05 0 0 4.28MB 44 页 10玖币
侵权投诉
In the realm of the hybrid brain: Human Brain and AI
Hoda Fares1, Margherita Ronchini1, Milad Zamani1, Hooman Farkhani1, Michela
Chiappalone2, Emre Neftci3, and Farshad Moradi1
1ICELab, IbrAIn center, Department of electrical and Computer Engineering, Aarhus University,
Aarhus, Denmark
2Department of Informatics, Bioengineering, Robotics and Systems Engineering (DIBRIS),
University of Genova, Genova, Italy
3 Peter Grünberg Institute, Forschungszentrum Jülich, Germany
*Correspondence:
Corresponding Authors: Farshad Moradi and Hoda Fares
Email: moradi@ece.au.dk, hfares@ece.au.dk
Keywords: a
Abstract
With the recent developments in neuroscience and engineering technology, it is now possible to
record brain signals and decode them. In parallel, a growing number of stimulation methods are being
utilized to modulate and influence brain activity. These advancements opened the door for innovative
neurotechnologies that directly interface with the human brain. Although current brain-computer
interface (BCI) technology is mainly focused on therapeutic outcomes, it already demonstrated its
efficiency as an assistive and rehabilitative technology for patients with severe motor impairments.
At the same time, artificial intelligence (AI) and machine learning (ML) have been recently used to
understand the enormous multimodal neural data and to decode brain signals. Beyond this progress,
interconnecting AI with advanced brain-computer interfaces in the form of implantable
neurotechnologies grant unique possibilities for the diagnosis, prediction, and treatment of
neurological and psychiatric disorders. In this context, we envision the development of a closed-loop
intelligent, low-power, and miniaturized neural interface that uses brain-inspired techniques to
process data from the brain; referred as Brain Inspired- Brain Computer Interface/Implant (BI-BCI).
Such a neural interface would offer access to deeper brain regions for a better understanding of brain’s
functions, thus improving BCIs operative stability and system’s efficiency. On one hand, brain
inspired-AI algorithms represented by spiking neural networks (SNNs) would be used to interpret the
multimodal neural signals in the BCI system. On the other hand, due to the ability of SNNs to capture
the rich dynamics of biological neurons and to represent and integrate different information
dimensions such as time, frequency, and phase, they would be used to model and encode complex
information processing in the brain and to provide feedback to the users. In this paper, we provide an
overview of different methods to interface with the brain and discuss the merger of AI with BCI, as
BI-BCI systems for present and future applications.
1 Neurotechnology: The future game changer
Neurodegenerative disorders such as Parkinson's Disease (PD), Epilepsy, Multiple Sclerosis (MS),
Alzheimer, and Dementia, are incurable and debilitating conditions caused by gradual damage or loss
of the nervous system structure and function. They lead to cognitive, sensory, and motor dysfunction.
As the world's population ages and life expectancy increases, age-related neurodegenerative diseases
are becoming more prevalent and the risk of being affected by them is increasing dramatically [2].
Such diseases are responsible for the greatest economic burden and influence the lives of millions of
people worldwide, for instance, in 2010, more than 179 million people in Europe were affected by
brain disorders with an associated bill of around 800 billion euros [3]. According to the Global Burden
of Disease Injuries, and Risk Factors Study (GBD) in 2016, neurological disorders were reported as
the top leading causes of disability in the globe with 11.6% Disability-Adjusted Life Years (DALYs)
(~276 million per year), and second leading cause of deaths after cardiovascular diseases with 16.5%
of all deaths (~9 million) [3]. A general summary of the most common neurological disorders, their
effects and economic burden is listed in Table 1. Currently, there is no effective therapeutics to cure
such disorders, except for some traditional pharmaceutical drugs that could reduce the symptoms
severity such as dopaminergic treatment for PD and movement disorders, cholinesterase for cognitive
disorders, anti-inflammatory and analgesic for neuronal infections and pain, antipsychotic for
dementia, etc.[4], [5]. To this end, a large body of research is focusing on establishing novel
therapeutic tools and strategies by targeting the nervous system, as in the case of Deep Brain
Stimulation (DBS) [6]–[8] , as alternative treatment to the traditional pharmaceutical approaches.
In the last 20 years, neurotechnologies aimed at interfacing the brain with machines and computers
(i.e., BMI/BCI Brain Machine Interface/ Brain- Computer Interface) emerged as interesting tools to
allow paralyzed people to communicate and interact with the external world. At the same time, they
started to be used to investigate brain functions in different experimental conditions.
Neurotechnologies or specifically neural interfaces cover any method or electronic device (e.g.,
electrodes, computers, robotic arm, etc.) that interface with the nervous system to monitor or alter
neural activity. They can either record and decode the brain signals into control commands or
electrically stimulate the brain to modulate its activity. Several neurotechnologies have been
developed in the past few decades which proved to be useful for both assistive and rehabilitative
applications, for example in cochlear implants for restoring hearing [9], retinal implants for restoring
vision [10], [11] , and brain-computer interfaces (BCIs) for brain-controlled applications [12] . More
recently, the advances in neuroscience and engineering technologies, along with the development of
Artificial intelligence (AI) and machine learning-related techniques have allowed neurotechnologies
to become intelligent for achieving a better performance [13]–[16].
Nowadays, researchers consider neurotechnologies to be the next game-changer for diagnosis,
treatment and even prediction of neurological and psychiatric disorders [13], [17], [18]. However,
most of the current ones are still limited to laboratories and their performance needs to be improved
so that they can be used in real life scenarios [18].
Table 1: Top leading neurodegenerative diseases based on world health organization (WHO) reports [2], [3], [19]
Neurodegenerative
Diseases
Facts and Symptoms
Percentages and economic Burden
Dementia and
Alzheimer’s disease
- Dementia causes symptoms that affect
memory, thinking, and social abilities
severely enough to interfere with a patient’s
daily life.
- Memory Loss, planning difficulties, mood
changes, personality changes, Confusion
about time and place.
- Over 50 million people worldwide were living with
dementia in 2020 (will double every 20 years).
- ~10 million new cases every year (one every 3 seconds).
- 7th leading cause of death.
- In 2018, it costed one trillion USD (it will be around two
trillion by 2030).
Parkinson’s disease (PD)
- Rigidity, postural disturbance, rest tremor,
slow movement, anosmia in early stages.
- 10 million patients affected globally (1.5x more likely
men than women)
- The prevalence ranges from 41 per 100,000 among people
in their thirties to more than 1,900 per 100,000 among
those who are over 80.
- In 2016, it caused 3.2 million DALYs and 211.96 deaths.
- In 2021, in the USA it costed 51.9 billion USD (double
previous estimates).
Multiple sclerosis (MS)
- Multiple sclerosis is a disease with
unpredictable symptoms that can also vary
- ~around 2.8 million people worldwide registered.
- Women four times more likely to have MS than men.
in intensity. Different symptoms can
manifest during relapses or attacks.
- Pain from spasticity, impaired ambulation,
depression, cognitive impairment, ataxia,
and tremor.
- Mean costs of MS ~37100 USD annually per patient with
moderate disease in EU.
Epilepsy
- Recurrent seizures, which are brief episodes
of involuntary movement that may involve a
part of the body or the entire body and are
sometimes accompanied by loss of
consciousness and control of bowel or
bladder function.
- ~ 50 million people worldwide have epilepsy (most
common neurological disease globally).
- Up to 70% of people living with epilepsy could be
seizure-free if properly diagnosed and treated.
- Premature death in people with epilepsy is up to three
times higher than for the general population.
- The estimated proportion of the general population with
active epilepsy (i.e., continuing seizures or with the need
for treatment) at a given time is between 4 and 10 per 1000
people.
The human brain is an extremely complex system and thus an active area of research for
neuroscientists and clinicians in designing treatments of (non)-neurological disorders, and for
engineers for its capability to perform complex tasks by means of ultra-energy-efficient computing.
Therefore, knowing how the brain works can be beneficial for both communities. In support of this,
vast resources have been assigned to study, model and map the brain and its fundamental mechanisms
along with neurotechnology development. The BRAIN initiative in 2013 supported by US
government, Brain/MINDS (Brain Mapping by Integrated Neurotechnologies for Disease Studies)
project launched in 2014 by Japan, and the Human Brain Project (HBP) funded by the European
commission (($703 million) are a few examples. In Dec 2020, the HBP launched its EBRAINS
platform, which grants access to datasets and digital tools for analysis and experiment conduction
[20]. Due to its high potential for treating neurological disorders, neurotechnology research has
significantly became an interesting attraction for industry in the past decade (e.g., © Neuralink [21],
©Paradromics [22], ©Synchron [23], ©Blackrock Neurotech [24], ©Neurable [25], ©Thync [26],
©Medtronic [27], ©kernel [28], etc.). For instance, ©NeuroPace developed a brain-responsive
neurostimulator called RNS System for treating adults with drug resistant focal epilepsy, using feature
thresholding over 4 channels to detect seizures [29], [30]. Also, ©Medtronic developed Percept PC
DBS system by that implements 4 Channels [31] and ©Neuralink developed a 1024 channel closed-
loop Brain Machine Interface (BMI) implantable chip integrating neural recoding, spike detection
circuitry while using external devices for motor intention decoding [32].
In this perspective paper, we aim to provide an overview of the current state of applied research in
neurotechnology including neural interfaces, neuroprostheses, BMIs/BCIs and surmise about future
developments and clinical application that may arise from it. Furthermore, we will delve into the co-
integration of AI-based processing and neural interfaces. The paper is structured as follows: section
2 presents a comprehensive synopsis about methodologies used to extract and transmit information
from and to the brain. Section 3 reviews the new generation of AI systems called spiking neural
networks or spiking neuromorphic architectures. Section 4 discusses the use of SNNs in
neurotechnology. Finally, the last section offers our closing remarks and our vision about merging
brain-inspired computing with neural interfaces to achieve Brain Inspired- Brain Computer
Interfaces/Implants (BI-BCIs) that would be the new generation of low-power, smart, and
miniaturized therapeutic devices for a wide range of neurological and psychiatric disorders.
In this context of this opinion, we use the “BCI” term for any technology that communicates directly
with the brain, either to extract information from it, or to feed information into it by means of brain
stimulation. Also, BCIs, or Brain-Computer Interfaces, are also commonly referred to as Brain-
Machine Interfaces and Neural Interfaces.
2 Neural Interfaces: The Brain Editors
2.1 Connecting Brain to Computers: Neural interfaces History and outlook
BCI is a general term for any technology that directly communicates with the brain and extracts
information from it either by observing its unperturbed electrical signal or by eliciting a measurable
neural response (evoked potential) through sensory stimulation [8][9] (Fig. 1). This terminology was
introduced by Vidal in 1970s, who first attempted to create a system capable of translating EEG
signals (i.e., Electroencephalography non-invasive method that records neural activity from the scalp)
into computer control signals [33]. Research applications of BCI technology have evolved
substantially over the past two decades [34]. Research and development of BCI technologies was
boosted by the technological advancement of microelectrode and single neuron recordings
technologies, both in rodents [35] and non-human primates [36], [37]. Researchers used electrode
arrays and implanted them in the parietal or motor cortex of patients with severe paralysis and
tetraplegia to perform skilled motor movements with a robotic arm [38]–[41]. BCIs could be used to
restore (e.g., unlock patients with locked-in syndrome), replace (e.g., BCI-controlled
neuroprosthesis), enhance (e.g., user experience enhancement through computer games), supplement
(e.g., VR, virtual reality, and AR, augmented reality, glasses), improve (e.g., lower limb rehabilitation
after stroke), and as a research tool (e.g., coding, and decoding brain activity with real-time feedback)
[34], [42], [43]. BCIs can record and decode cortical activity while performing or imagining
performing a task. The neural signals related to the intended movement can be transformed into visual
[44], auditory [45]–[47], or haptic feedback of the movement [34], [48]. Figure. 1 illustrates the
generalized schematic for BCIs and common state-of-art (SoA) applications.
BCIs can be classified based on the way they interact with the brain. First category is the active BCIs
that either use the users consciously induced brain activity such as Motor Imagery (MI) [46], [49],
[50] or induced brain activity by external stimuli (e.g., visual, auditory, or somatosensory stimuli)
[34], [49], [51], [52] . While the second category called passive BCIs decode brain’s unconscious
psychological states and do not require an active participation from the user [53], [54]. They have
been used to monitor users’ cognitive states such as intentions, emotional states, situational
interpretations [45], [54], and drowsiness [55], [56].
Figure 1 Generalized schematic for Bidirectional brain computer Interface (BBCI). The produced Brain signals are recorded from the scalp,
cortical surface or from within the brain by electrodes. These signals are processed to extract the correlated features with user’s intentions. The
extracted features are translated into commands to control / actuate a wide range of applications. (Could be used to control devices, artificial
limbs, or obtain knowledge of user’s intentions). Then, sensory information is fed back to the user either invasively or non-invasively.
Signal Processing
Pre-processing Feature Extraction Classification
Signal
Acquisition
Commands
Actuate/Control
User Applications
Feedback
BCI Applications Neuroprosthesis
Spelling Device
Wheelchair Control
Restore lost
Communication
Rehabilitation
Spelling Device
Sensory-Motor
Restoration
User
Bidirectional Brain-computer Interface System
2.2 From the brain to external devices: Recording and Decoding
2.2.1 Recording techniques
Several techniques are employed to gather metabolic and electrophysiological signals from the brain,
each offering distinct temporal and spatial resolution. Intracellular recordings measure the voltage
across the cell membrane of a single neuron by placing electrodes inside and outside the membrane,
also they capture the sub-threshold variations from resting potential. Extracellular recordings capture
the summation of signals by nearby neurons, and they provide lower neural signal amplitudes in
comparison to intracellular recordings, but they cover larger neural areas [57]. Based on electrode
location, extracellular recordings techniques can be categorized into either invasive or non-invasive
methods [12], [34]. The most relevant non-invasive and invasive recording techniques respectively
and their BCI applicability are summarized in Table 2 and 3.
Non-invasive methods used for neural signals recording comprise electroencephalography (EEG)
[31], magnetoencephalography (MEG) [58], [59], and metabolic signals recorded either by
functional near-infrared spectroscopy (fNIRS) [52], [60], or functional magnetic resonance
imaging (fMRI) [61].I EEG is the most employed technique in clinical setups for diagnosis purposes
due to its non-invasive nature and ease of use. However, this technique captures only collective
information from the top cortical layers of the brain, and it suffers from low spatial resolution, poor
contact between the electrode and the scalp and low signal quality [62]. Wet EEG electrodes are
typically made of metals and gels and mounted in elastic caps to enhance the signal quality. Dry
electrodes (i.e., without gels) are more favorable and have a comparable performance with wet
electrodes, yet they are less robust to moving artifacts and show higher electrode-tissue impedance
[63], [64]. To address these challenges, active electrodes with integrated preamplifiers have been
developed, also new materials has been used to design EEG electrodes such as polymer foam
electrodes, soft conductive textiles electrodes, etc. fMRI, as another non-invasive the uses blood-
oxygen-level-dependent (BOLD) signals that reflect changes in deoxyhemoglobin driven by
localized changes in brain blood flow and blood oxygenation, which are coupled to underlying
neuronal activity by a process termed neurovascular coupling. But it is more expensive method in
clinical setups, offers a much higher spatial resolution metabolic signals (~1mm) and is more sensitive
to subcortical regions than electrophysiological signals. fMRI is heavily used in cognitive research
[65]. Researchers were able to reconstruct perceived visual images, just by analyzing fMRI signals
collected from visual cortex [66]. With similar approaches, it has been demonstrated that patients in
a vegetative or minimally conscious state understand and respond to instructions [67], [68] (Table 2).
Alternatively, invasive methods such as electrocorticography (ECoG or μEoG) [57], [69], [70] and
intracortical recordings (IR) [71]–[73] provide higher signal-noise ratio and higher-frequency
signal bands, as well as better localization of brain activity as they enable more direct interaction with
the brain. For instance, flexible μECoG electrodes have pushed spatial resolution down to 1mm even
sub-mm range [74] unlike conventional ECoG electrodes which have a pitch of around 1 cm [75].
Transistor multiplexed ECoG arrays managed to increase the electrode density and channel count and
reduce the area for routing wires [76], [77]. Both ECoG and μECoG are used in preclinical and
clinical research settings [78], [79]. Lately, bundled arrays of microwires were used to interface with
up to 1 million neurons through a neural input-output bus (NIOB) funded by DARPA [80]. Despite
the improved performance in spike sorting and mechanical stability offered by microwires, this
method still faces challenges in signal attenuation and cross talk. To overcome these limitations,
silicon-based needle shaped microelectrodes enabling multisite recording were proposed [81]–[83] .
摘要：

Intherealmofthehybridbrain:HumanBrainandAIHodaFares1,MargheritaRonchini1,MiladZamani1,HoomanFarkhani1,MichelaChiappalone2,EmreNeftci3,andFarshadMoradi11ICELab,IbrAIncenter,DepartmentofelectricalandComputerEngineering,AarhusUniversity,Aarhus,Denmark2DepartmentofInformatics,Bioengineering,RoboticsandS...

展开>> 收起<<
In the realm of the hybrid brain Human Brain and AI Hoda Fares1 Margherita Ronchini1 Milad Zamani1 Hooman Farkhani1 Mich ela Chia ppalone2 Emre Neft ci3 and Farshad Moradi1.pdf

共44页,预览5页

还剩页未读， 继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

相关推荐

更多
立即下载
分类：图书资源 价格：10玖币 属性：44 页 大小：4.28MB 格式：PDF 时间：2025-05-05

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务

作者详情

相关内容

更多

热门标签

人际关系 动力学连接体 力的合成 配电装置 高考理综 全宋诗作者索引 公务员考试
/ 44
评分 收藏
分享
立即下载
客服
关注