Artificial Intelligence in Neurological Diagnosis & Management | 18th Neurology Certificate Course

00:42:04
https://www.youtube.com/watch?v=Kmrt4T-hwVM

الملخص

TLDRDr. Jeanette's presentation at the neurology certificate course addresses the rapidly evolving intersection of artificial intelligence (AI) and neurology. She emphasizes how AI can significantly impact the diagnosis and treatment of neurological disorders, particularly in stroke care and epilepsy. The lecture highlights the urgent need for healthcare professionals to understand AI as an essential part of their practice. Key topics covered include the acceleration of AI research publications, the importance of high-quality data, and the potential of AI to reduce latency in care delivery, ultimately improving patient outcomes. Dr. Jeanette also discusses how AI can develop new biomarkers for diseases like Alzheimer's disease and stresses the need for continuous learning environments in AI applications within medicine. Overall, she advocates for the integration of AI in clinical settings to enhance decision-making and prediction capabilities while maintaining the crucial role of physicians.

الوجبات الجاهزة

  • 🤖 AI is revolutionizing neurology and clinical practice.
  • 📈 AI publications are increasing rapidly, led by tech companies.
  • 🧠 AI can reduce treatment latency in stroke care.
  • 🔬 High-quality data is essential for effective AI applications.
  • 💡 Understanding AI is now as critical as learning medical biochemistry.

الجدول الزمني

  • 00:00:00 - 00:05:00

    Jeanette, a neurocritical care specialist, discusses the role of artificial intelligence (AI) in neurological disorders during the 18th neurology certificate course at Aahan University Hospital. She shares her background and ongoing work in the field, emphasizing the importance of understanding clinically relevant AI concepts.

  • 00:05:00 - 00:10:00

    AI is rapidly changing healthcare by affecting algorithms that can influence billions of people globally. Jeanette highlights the significant increase in AI publications, particularly in neurology, stressing that academia is lagging behind tech companies like Facebook and Google.

  • 00:10:00 - 00:15:00

    The rise in AI publications correlates with the availability of high-quality data, particularly in neurodegenerative disorders like Alzheimer's and Parkinson's. Furthermore, Jeanette cites the increasing relevance of AI in stroke care, emphasizing the importance of real-world applications in contrast to academia.

  • 00:15:00 - 00:20:00

    AI development in healthcare is accelerating due to reduced costs of compute power and faster data storage. Jeanette notes that understanding AI concepts becomes essential for healthcare providers to improve treatment quality and reduce latency in care delivery, especially for conditions like stroke.

  • 00:20:00 - 00:25:00

    Jeanette emphasizes that it is crucial to understand AI as a technology enabler rather than just a tool. She stresses the need for medical professionals to learn about AI, likening its importance to foundational medical knowledge and asserting that its application will impact patient care significantly.

  • 00:25:00 - 00:30:00

    There are fundamental differences in reasoning methods between AI and traditional methods. AI, particularly machine learning and deep learning, relies on patterns and examples rather than rules and instructions, leading to more efficient hypotheses and discoveries in clinical practice.

  • 00:30:00 - 00:35:00

    Statistics and machine learning serve different purposes; statistics identify relationships while AI focuses on predictions. Jeanette underscores the shift from reactive disease management to proactive prediction and prevention through AI applications in healthcare.

  • 00:35:00 - 00:42:04

    Artificial intelligence is a critical tool that, while not replacing physicians, will aid in augmenting their capabilities. Jeanette forecasts a future where physicians using AI will outperform those who do not, advocating for an integration of AI in daily medical practices.

اعرض المزيد

الخريطة الذهنية

فيديو أسئلة وأجوبة

  • What is the main focus of Dr. Jeanette's presentation?

    The focus is on the impact of artificial intelligence in neurology, specifically for stroke and epilepsy.

  • How has the interest in AI publications changed recently?

    Publications in AI have been increasing rapidly, particularly from technology companies like Facebook and Google.

  • What are the two major applications of AI in neurology discussed?

    Stroke management and the development of biomarkers for diseases like Alzheimer's.

  • Why is AI considered a technology enabler?

    AI is seen as a general purpose technology that can be applied across various fields, not limited to one specific area.

  • How does AI help reduce latency in stroke care?

    AI applications in stroke care can decrease the time to treatment, potentially saving lives and reducing disability.

عرض المزيد من ملخصات الفيديو

احصل على وصول فوري إلى ملخصات فيديو YouTube المجانية المدعومة بالذكاء الاصطناعي!
الترجمات
en
التمرير التلقائي:
  • 00:00:00
    hi this is jeanette neurocritical care
  • 00:00:01
    stroke and epilepsy specialist today i'm
  • 00:00:03
    going to be presenting an artificial
  • 00:00:04
    intelligence and neurological disorders
  • 00:00:06
    i'm very grateful to be invited to speak
  • 00:00:09
    on the prestigious course which is the
  • 00:00:11
    18th neurology certificate course from
  • 00:00:14
    aahan university hospital being medical
  • 00:00:17
    advisor consultation for many different
  • 00:00:19
    companies and on artificial intelligence
  • 00:00:21
    that i can't list over here but these
  • 00:00:22
    are generally where my current positions
  • 00:00:25
    are in terms of
  • 00:00:27
    employment and contraction and some of
  • 00:00:29
    the independent work that i'm doing in
  • 00:00:31
    the field itself
  • 00:00:32
    now this is a very big topic and i
  • 00:00:35
    cannot boil an ocean
  • 00:00:37
    what that term means is basically that
  • 00:00:38
    this is a vast ocean in which we have to
  • 00:00:40
    get into so the objective is basically
  • 00:00:43
    to get as much information as you need
  • 00:00:45
    in terms of clinically relevant
  • 00:00:47
    artificial intelligence and getting the
  • 00:00:49
    basic concepts around it and then you
  • 00:00:51
    will improve upon it more and more your
  • 00:00:53
    knowledge in terms of looking at our
  • 00:00:55
    artificial intelligence articles youtube
  • 00:00:57
    videos etc and going from there my
  • 00:01:00
    articles my videos are more and more
  • 00:01:02
    going to be concentrating on artificial
  • 00:01:04
    intelligence so make sure you follow
  • 00:01:06
    me in terms of my youtube channel my
  • 00:01:09
    blog and of course neurology pocketbook
  • 00:01:11
    as well
  • 00:01:12
    so here's the issue
  • 00:01:14
    change was always changing right it's
  • 00:01:16
    never been that issue right i mean at
  • 00:01:18
    the end of the day the
  • 00:01:20
    acceleration of change itself has been
  • 00:01:22
    accelerating for the past decade what
  • 00:01:24
    artificial intelligence is that it
  • 00:01:26
    changed from logarithmic to algorithmic
  • 00:01:29
    change what's the difference the
  • 00:01:30
    difference is that if there's an
  • 00:01:32
    algorithmic change let's say in terms of
  • 00:01:33
    youtube algorithm it affects two billion
  • 00:01:36
    people tick tock three billion people
  • 00:01:38
    facebook three and a half billion people
  • 00:01:39
    it's ridiculous the amount of change
  • 00:01:42
    that you can produce when you change an
  • 00:01:44
    algorithm i mean we always have heard
  • 00:01:46
    this right i mean
  • 00:01:50
    this has been an alumni balls code that
  • 00:01:52
    we keep quoting in pakistan especially
  • 00:01:55
    for a long long time but believe it or
  • 00:01:58
    not the change itself has changed
  • 00:02:01
    interestingly this uh 1922 22 exactly a
  • 00:02:05
    hundred years ago allama iqbal was given
  • 00:02:07
    the title of sir so interestingly we
  • 00:02:09
    have been a 100 year journey and we are
  • 00:02:12
    still talking about changes change and
  • 00:02:14
    believe it or not this time it's very
  • 00:02:16
    significant
  • 00:02:18
    now publications in artificial
  • 00:02:20
    intelligence have been increasing at an
  • 00:02:21
    alarming rate as a matter of fact the
  • 00:02:23
    peer-reviewed applications there were
  • 00:02:25
    publications that were coming in
  • 00:02:26
    artificial intelligence is on a rise
  • 00:02:28
    constantly and more importantly start
  • 00:02:30
    just coming from the academics believe
  • 00:02:32
    it or not more and more publications
  • 00:02:34
    that i read personally are coming from
  • 00:02:36
    facebook and google because those are
  • 00:02:38
    the technology papers that i have to
  • 00:02:39
    make myself aware of because more and
  • 00:02:41
    more changes that these companies are
  • 00:02:43
    coming up with and they are doing real
  • 00:02:46
    good publication and sometimes it's open
  • 00:02:48
    source open access so first of all i
  • 00:02:50
    need to know them personally but just
  • 00:02:52
    letting you know that academia is truly
  • 00:02:54
    falling behind in terms of
  • 00:02:56
    research publications and artificial
  • 00:02:58
    intelligence so if you want to follow
  • 00:03:00
    that technological curve you need to
  • 00:03:02
    follow corporate publications as well as
  • 00:03:04
    academic publications as well
  • 00:03:06
    as far as neurology is concerned as you
  • 00:03:08
    can see every single year there is more
  • 00:03:10
    and more increase in terms of
  • 00:03:11
    publications and neurology these are
  • 00:03:13
    only specific clinical application
  • 00:03:15
    neurology publications that we are
  • 00:03:16
    talking about and you can see that from
  • 00:03:18
    2018 to 2020 there is clearly doubling
  • 00:03:21
    every year in terms of publications and
  • 00:03:23
    neurology is concerned all the other
  • 00:03:25
    places as far as you know
  • 00:03:28
    sub-specialties are concerned there has
  • 00:03:30
    been publications general publications
  • 00:03:32
    radiology is one of the top
  • 00:03:34
    publications where you would see
  • 00:03:36
    artificial intelligence being applied to
  • 00:03:39
    however cardiology and neurology are
  • 00:03:40
    neck to neck and if you see covet 19 of
  • 00:03:43
    course was a little higher in terms of
  • 00:03:45
    publications which is probably going to
  • 00:03:47
    fall off now given that we are going
  • 00:03:50
    over the hump of that so artificial
  • 00:03:52
    intelligence application has been used
  • 00:03:54
    in every freaking sub-specialty and
  • 00:03:57
    neurology is not behind as a matter of
  • 00:03:59
    fact neurology is right up at the level
  • 00:04:01
    of cardiology in terms of publications
  • 00:04:03
    as well
  • 00:04:04
    now the publications are divided into
  • 00:04:06
    five major categories and the issue here
  • 00:04:09
    is that the why is that those areas
  • 00:04:12
    being concentrated on it has nothing to
  • 00:04:14
    do with clinical application it has
  • 00:04:16
    everything to do with the availability
  • 00:04:18
    of high quality data
  • 00:04:20
    look alzheimer's parkinson's which are
  • 00:04:23
    the neurodegenerative and dementia
  • 00:04:25
    disorders have had multiple years of
  • 00:04:27
    funding from the government in terms of
  • 00:04:29
    fda nih and ihss and inds all these
  • 00:04:33
    major federal funding has been going
  • 00:04:35
    into dementia because alzheimer's has
  • 00:04:37
    been a big problem when we need newer
  • 00:04:39
    newer solutions
  • 00:04:41
    all those data that has been collected
  • 00:04:42
    in last 10 years from imaging biomarkers
  • 00:04:45
    everything can be churned up through
  • 00:04:47
    artificial intelligence better faster
  • 00:04:49
    because we have availability of high
  • 00:04:50
    quality data however the application the
  • 00:04:54
    true world application in artificial
  • 00:04:56
    intelligence is in stroke that's we're
  • 00:04:58
    going to come into in the next few
  • 00:05:00
    slides so you have to distinguish
  • 00:05:01
    between publications in terms of what
  • 00:05:04
    academia is moving towards and real life
  • 00:05:06
    applications so the reasoning behind it
  • 00:05:09
    is not just the publications itself the
  • 00:05:10
    reasoning behind it is basically
  • 00:05:12
    availability of high quality data
  • 00:05:14
    so
  • 00:05:16
    what it tells me everything is that ai
  • 00:05:18
    is developing super fast as a matter of
  • 00:05:20
    fact when you're going to look at it in
  • 00:05:21
    terms of both technological aspects and
  • 00:05:23
    in terms of clinical applications we are
  • 00:05:26
    seeing more and more applications that
  • 00:05:28
    are getting fda
  • 00:05:29
    i'm not going to use the word approval
  • 00:05:31
    so that's another big thing because fda
  • 00:05:33
    granted fda access fda listed and fda
  • 00:05:36
    approved however for now we're going to
  • 00:05:38
    see that ai is developing fast its
  • 00:05:41
    adoption has become faster as a matter
  • 00:05:43
    of fact more and more industry leaders
  • 00:05:45
    and more and more experts are
  • 00:05:48
    proposing that especially in healthcare
  • 00:05:50
    artificial intelligence should be the
  • 00:05:52
    way to go especially with hims22 if you
  • 00:05:54
    look at the news that came from there
  • 00:05:56
    more and more leaders are going to be
  • 00:05:58
    open to adopting artificial intelligence
  • 00:06:01
    now why now i think there are multiple
  • 00:06:03
    reasons and this is a whole chart and we
  • 00:06:05
    can go into but given the time i'm going
  • 00:06:08
    to just talk about two things number one
  • 00:06:10
    is
  • 00:06:10
    the key issue is compute power it has
  • 00:06:13
    become significantly cheaper and it's
  • 00:06:16
    significantly accessible in terms of
  • 00:06:18
    compute power because we can do compute
  • 00:06:20
    over the cloud and even our devices are
  • 00:06:22
    getting much and much powerful in terms
  • 00:06:24
    of compute second big thing is data and
  • 00:06:26
    not just data in terms of slow hard
  • 00:06:29
    drive spinning drives data hard
  • 00:06:32
    solid state drive which is the flash
  • 00:06:34
    drives and the reason behind is that we
  • 00:06:36
    need faster drives sometimes so that the
  • 00:06:39
    data can go in and out pretty fast
  • 00:06:41
    within the computer engines that is
  • 00:06:43
    decreasing in cost so two major storage
  • 00:06:45
    and compute power is decreasing in cost
  • 00:06:47
    significantly so the accessibility for
  • 00:06:50
    the barrier to entry has gone
  • 00:06:51
    significantly low for these studies to
  • 00:06:53
    come up and therefore more and more
  • 00:06:55
    studies are being done and more and more
  • 00:06:57
    applications are being done on
  • 00:06:59
    artificial intelligence before we get
  • 00:07:01
    into it i always always remember that we
  • 00:07:03
    need to talk about why is it important
  • 00:07:06
    you know you can already apply
  • 00:07:07
    artificial intelligence in any data
  • 00:07:09
    right i mean let's say i'm gonna do uh
  • 00:07:11
    pull out some of the tweet data from
  • 00:07:13
    myself and then put artificial
  • 00:07:15
    intelligence great right i mean we can
  • 00:07:16
    do that but the more important thing is
  • 00:07:18
    to always find out why we want to do
  • 00:07:20
    something especially in terms of
  • 00:07:22
    technological applications so the key
  • 00:07:25
    issue with a neurology is what
  • 00:07:27
    latency of care time is brain so one of
  • 00:07:30
    the things that we want to apply
  • 00:07:31
    artificial intelligence in situations
  • 00:07:33
    where we can decrease the time for
  • 00:07:35
    treatment stroke hemorrhage subarachnoid
  • 00:07:38
    hemorrhage etc the second big issue is
  • 00:07:40
    that we don't have a troponin like i
  • 00:07:41
    mean that would be fantastic right if
  • 00:07:43
    you can get a blood test for alzheimer's
  • 00:07:45
    we can't if you had that that would be
  • 00:07:47
    that would solve so many issues because
  • 00:07:48
    we can have early detection of
  • 00:07:50
    alzheimer's and then early intervention
  • 00:07:52
    of the alzheimer's we can't do that so
  • 00:07:54
    we need new
  • 00:07:55
    significantly new biomarkers in terms of
  • 00:07:58
    imaging in terms of mci clinical
  • 00:08:00
    biomarkers and in terms of genetic
  • 00:08:02
    biomarkers and meaning we probably need
  • 00:08:03
    to combine them so the question is that
  • 00:08:05
    we need to make sure that we have new
  • 00:08:07
    forms of biomarkers that could only be
  • 00:08:10
    achieved by large data
  • 00:08:12
    churned by artificial intelligence we're
  • 00:08:14
    going to continue to have lack of
  • 00:08:15
    neurologist i mean pakistan has a
  • 00:08:17
    significant lack of neurologist then on
  • 00:08:19
    top of that we also have a long tail of
  • 00:08:21
    these patients that are chronically
  • 00:08:22
    going to live with neurological
  • 00:08:24
    disorders either parkinson's alzheimer's
  • 00:08:27
    stroke etc epilepsy all of these things
  • 00:08:29
    we're going to significantly see a
  • 00:08:31
    growing population of older and older
  • 00:08:33
    people around the world and they are all
  • 00:08:36
    going to get you know basically
  • 00:08:39
    neurological problems and we need to
  • 00:08:40
    help them and this is what i keep
  • 00:08:42
    telling cardiology of my cardiology
  • 00:08:44
    folks is that thank you for making sure
  • 00:08:46
    that people are alive so i can take care
  • 00:08:47
    of their neurological problems because
  • 00:08:49
    that was the last
  • 00:08:50
    century to be honest with you where
  • 00:08:52
    cardiology was at the boom in which
  • 00:08:54
    heart problems were the issue when
  • 00:08:55
    people were constantly getting heart
  • 00:08:57
    attacks and dying from it but in their
  • 00:08:58
    lot now they're living well
  • 00:09:01
    and unfortunately being afflicted by
  • 00:09:03
    neurological disorders
  • 00:09:05
    now before i get into the direct
  • 00:09:07
    applications of neurology let me just
  • 00:09:09
    give you some conceptual basics of
  • 00:09:10
    artificial intelligence first and
  • 00:09:12
    foremost you need to understand that
  • 00:09:13
    artificial intelligence is not a
  • 00:09:15
    technology it's a technology enabler
  • 00:09:18
    what that means is that any type of data
  • 00:09:20
    that you're going to generate in either
  • 00:09:22
    through blockchain dna sequences
  • 00:09:25
    internet of things healthcare iot
  • 00:09:27
    autonomous mobiles robotics any
  • 00:09:29
    situation in which you're going to
  • 00:09:31
    generate data and you need to have run
  • 00:09:33
    assimilations of multiple possibilities
  • 00:09:36
    you're going to need artificial
  • 00:09:37
    intelligence
  • 00:09:38
    so this is considered what we call
  • 00:09:40
    general purpose technology it is not
  • 00:09:42
    considered one technology so therefore
  • 00:09:45
    when you have a technology like general
  • 00:09:46
    purpose technology like artificial
  • 00:09:48
    intelligence you do not compare it with
  • 00:09:50
    just dna sequences because that's one
  • 00:09:53
    part of technology that cannot be
  • 00:09:54
    applied in finance but artificial
  • 00:09:57
    intelligence can be applied in finance
  • 00:09:59
    healthcare economics education
  • 00:10:01
    everything so therefore artificial
  • 00:10:03
    intelligence is to be compared with
  • 00:10:05
    electricity so its artificial
  • 00:10:07
    intelligence is as important as
  • 00:10:09
    electricity in terms of its impact
  • 00:10:12
    because it's a general purpose
  • 00:10:14
    technology and that's why there's such a
  • 00:10:16
    huge buzz and everything i mean there's
  • 00:10:18
    a lot of buzz so don't get into it but
  • 00:10:19
    at the end of the day that's where the
  • 00:10:21
    real difference is in terms of just
  • 00:10:24
    being a technology or being a technology
  • 00:10:26
    enabler
  • 00:10:27
    now
  • 00:10:28
    second thing that you have to understand
  • 00:10:30
    is like the first question why do i need
  • 00:10:31
    to know ai i'm going to give you many
  • 00:10:33
    examples of artificial intelligence
  • 00:10:34
    being used in clinical impact
  • 00:10:37
    exactly the same way that you have
  • 00:10:39
    learned
  • 00:10:40
    mechanism of actions of drugs all your
  • 00:10:42
    life right like this is topo mermaid
  • 00:10:45
    this is kapra this is a mechanism action
  • 00:10:47
    this is the half-life this is what that
  • 00:10:49
    is a tool that you use to help your
  • 00:10:51
    patients exactly in the same way you're
  • 00:10:54
    going to be using artificial
  • 00:10:55
    intelligence in the future and exactly
  • 00:10:57
    in the same way you need to understand
  • 00:10:58
    some basic some mechanism of action for
  • 00:11:01
    artificial intelligence to make sure
  • 00:11:03
    that you are appropriately applying
  • 00:11:05
    those tools for your patient care
  • 00:11:07
    so do not think that artificial
  • 00:11:08
    intelligence is something you do not
  • 00:11:10
    need to know it is now as important as
  • 00:11:12
    learning biochemistry in pre-med and
  • 00:11:15
    then biochemistry in med school or
  • 00:11:16
    anatomy or physiology there should be
  • 00:11:18
    courses on data sciences and artificial
  • 00:11:20
    intelligence in medicine this is not the
  • 00:11:22
    time i can give you some more big
  • 00:11:24
    references on where
  • 00:11:25
    at least universities in u.s are going
  • 00:11:27
    if you want to check out northwestern
  • 00:11:28
    university just added a complete course
  • 00:11:30
    on data sciences and artificial
  • 00:11:32
    intelligence in their medical school
  • 00:11:33
    curriculum the reason is this you are
  • 00:11:36
    going to be taking care of these
  • 00:11:37
    patients using these tools so you need
  • 00:11:40
    to make sure you understand some basics
  • 00:11:42
    at least enough that you
  • 00:11:44
    currently know what the mechanism of
  • 00:11:46
    action for some medicines are and choose
  • 00:11:48
    accordingly which medication you're
  • 00:11:49
    going to use and which you're not going
  • 00:11:50
    to use exactly in the same way you have
  • 00:11:52
    to start using artificial intelligence
  • 00:11:54
    which software stack you're going to use
  • 00:11:56
    and which you're not going to use to
  • 00:11:57
    help this particular patient so that's
  • 00:12:00
    why it is extremely important and this
  • 00:12:01
    is your role as a physician to
  • 00:12:03
    understand this in the future and the
  • 00:12:05
    future is not that far as i told you
  • 00:12:07
    before now when you're talking about
  • 00:12:08
    artificial intelligence what we're
  • 00:12:09
    talking about is machine learning and
  • 00:12:11
    that's a lot of good term that i use the
  • 00:12:13
    word artificial intelligence because
  • 00:12:14
    that's a general purpose technology and
  • 00:12:16
    there are different ways to globally
  • 00:12:17
    tell something and the other way is not
  • 00:12:19
    to say generally the best term to use is
  • 00:12:21
    machine learning and the one of the key
  • 00:12:23
    issues with the machine one of the
  • 00:12:25
    subsets in machine learning is deep
  • 00:12:26
    learning so let's talk about the the one
  • 00:12:29
    two things we're going to be talking
  • 00:12:30
    about is machine learning and deep
  • 00:12:31
    learning machine learning is basically
  • 00:12:33
    that when a machine basically learns
  • 00:12:35
    from examples and deep learning is
  • 00:12:37
    basically in which we are not giving too
  • 00:12:39
    much labeled data and more and more
  • 00:12:41
    excess exceptionally large amount of
  • 00:12:44
    data is added in terms of deep learning
  • 00:12:47
    and therefore when you are trying to
  • 00:12:48
    explain so colloquially we use the term
  • 00:12:51
    artificial intelligence which is to be
  • 00:12:52
    honest with you not the right term the
  • 00:12:54
    term we should be using is machine
  • 00:12:55
    learning and whenever we are referring
  • 00:12:57
    to artificial intelligence especially in
  • 00:12:58
    healthcare we're talking about machine
  • 00:13:00
    learning
  • 00:13:00
    machine learning has another subset
  • 00:13:02
    called deep learning and deep learning
  • 00:13:04
    is basically when you have more higher
  • 00:13:06
    forms of learning that are compute
  • 00:13:08
    intensive and data intensive in that
  • 00:13:10
    regard we have to see that when you are
  • 00:13:13
    moving from machine learning to deep
  • 00:13:14
    learning there's massively increasing
  • 00:13:17
    amounts of requirements for data and
  • 00:13:19
    massively increasing requirements of
  • 00:13:21
    compute power however we lose one key
  • 00:13:24
    issue we lose explainability because in
  • 00:13:27
    machine learning we are giving them at
  • 00:13:28
    least some sort of rules to go back and
  • 00:13:31
    then we can explain that how that result
  • 00:13:33
    was coming up to but in deep learning
  • 00:13:35
    that is not the case let me explain
  • 00:13:37
    further
  • 00:13:38
    so what is the difference right i mean
  • 00:13:41
    what's the freaking hype about in terms
  • 00:13:43
    of artificial intelligence we have been
  • 00:13:45
    you know have human intelligence for
  • 00:13:46
    years and years right so we have been
  • 00:13:48
    programming for years like it's been
  • 00:13:50
    approximately 50 years of coding and
  • 00:13:53
    programming that has been going on how
  • 00:13:54
    do we do that human coders said they
  • 00:13:57
    put in spit out a lot of instructions
  • 00:13:59
    and rules that rules and instructions
  • 00:14:01
    basically create a program
  • 00:14:03
    something like this in which there's
  • 00:14:04
    video uploaded on the youtube etc and
  • 00:14:07
    those instructions are basically
  • 00:14:08
    packaged and then go into it what's the
  • 00:14:11
    difference is that then we no longer
  • 00:14:13
    want human to code data codes we
  • 00:14:17
    basically do not give them instructions
  • 00:14:19
    we give them examples
  • 00:14:21
    let me rephrase it
  • 00:14:23
    you know there are ten thousand cat
  • 00:14:25
    pictures and we can tell them okay do
  • 00:14:27
    something about it then the the computer
  • 00:14:29
    will never learn because no matter how
  • 00:14:30
    many times you have instructions to
  • 00:14:33
    divide that from between
  • 00:14:34
    dog and cat it just never works this is
  • 00:14:37
    a completely new paradigm in which you
  • 00:14:39
    do not give instructions to the page to
  • 00:14:42
    the computer you give okay here are a
  • 00:14:44
    million pictures this is the dog this is
  • 00:14:47
    the cat and then learn from it and
  • 00:14:49
    here's another 10 000 more where you're
  • 00:14:51
    gonna be able to distinguish between the
  • 00:14:54
    two sets of data and test them we'll
  • 00:14:56
    come to that in a minute so the key
  • 00:14:59
    difference is human codes to data code
  • 00:15:02
    instructions to examples we give tons
  • 00:15:05
    and tons of examples to the computer and
  • 00:15:08
    that's how it learns and that's to be
  • 00:15:10
    honest with you is the basis of human
  • 00:15:11
    learning as well another big difference
  • 00:15:13
    is that that reasoning and artificial
  • 00:15:15
    intelligence look what we have been
  • 00:15:17
    taught about is deductive reasoning
  • 00:15:19
    that's how we run virtually clinical
  • 00:15:21
    trials that's how we move on in terms of
  • 00:15:24
    scientific discovery scientific
  • 00:15:26
    validation what do we do we have a
  • 00:15:28
    theory that i don't know tevera made uh
  • 00:15:30
    helps with the headache then we create
  • 00:15:32
    that into a hypothesis then we create
  • 00:15:35
    virtual clinical trials divide patients
  • 00:15:36
    into two arms we do observations and
  • 00:15:38
    then we confirm that that if our
  • 00:15:41
    hypothesis was correct or wrong
  • 00:15:44
    this is our current methods of thinking
  • 00:15:46
    and this is what we call aristotle or
  • 00:15:48
    deductive thinking
  • 00:15:50
    sherlock or inductive thinking has been
  • 00:15:52
    that we don't care we just say there are
  • 00:15:56
    tons and tons of millions of you know
  • 00:15:57
    records and and here and then just find
  • 00:16:00
    out patterns in it and then see with the
  • 00:16:02
    patterns if we can create some more
  • 00:16:04
    hypothesis and then we create theory
  • 00:16:07
    so for example there are a million
  • 00:16:08
    epilepsy patients who are on topiramate
  • 00:16:11
    and then we saw that they also had some
  • 00:16:12
    10 000 of them had migraine and we saw
  • 00:16:15
    that those patients were on top pyramid
  • 00:16:16
    had a decreased headache frequency
  • 00:16:19
    because they were on the
  • 00:16:21
    that's artificial intelligence because
  • 00:16:23
    you're putting in a lot of data which
  • 00:16:25
    there's pattern recognition and it gives
  • 00:16:27
    you a theory rather than the other way
  • 00:16:29
    around so artificial intelligence is
  • 00:16:31
    inductive reasoning artificial
  • 00:16:33
    intelligence starts with observation
  • 00:16:35
    examples in terms of deductive reasoning
  • 00:16:38
    it starts with theory and confirmation
  • 00:16:40
    so that's a completely different way of
  • 00:16:41
    looking at things and then you will have
  • 00:16:44
    to do a complete clinical trials as well
  • 00:16:46
    but what ai can do is get you there
  • 00:16:48
    faster can get you there smarter and
  • 00:16:50
    that's where drug discoveries are being
  • 00:16:52
    done through artificial intelligence
  • 00:16:54
    that's exactly the pathway that they go
  • 00:16:55
    into which was a different pathway for
  • 00:16:57
    what we were doing before because we
  • 00:16:59
    were now able to select compounds much
  • 00:17:01
    faster much easier much smarter to do
  • 00:17:04
    for clinical trials through the
  • 00:17:05
    inductive reasoning process which uses
  • 00:17:08
    which is based artificially intelligence
  • 00:17:10
    is based out of
  • 00:17:11
    and this is what i was talking about
  • 00:17:12
    practically skipping when you have an
  • 00:17:14
    original data format you turn it into a
  • 00:17:16
    training data set into a testing data
  • 00:17:18
    sometimes you do have a validation data
  • 00:17:20
    set as well in between to hypertune your
  • 00:17:23
    parameters but either way the logic
  • 00:17:25
    remains the same that you take a tons
  • 00:17:27
    and tons amount of data separate it down
  • 00:17:29
    into training data and say okay these
  • 00:17:30
    are label data these are cats these are
  • 00:17:32
    dogs and then these are unlabeled data
  • 00:17:34
    in which you know i'm going to test how
  • 00:17:36
    do you perform in identifying those
  • 00:17:39
    images the other thing you need to know
  • 00:17:40
    is what's the real difference between a
  • 00:17:42
    statistics and machine learning you
  • 00:17:44
    understood the in reasoning behind
  • 00:17:45
    artificial intelligences which is
  • 00:17:47
    different than what we have been doing
  • 00:17:48
    and that's why it's harder especially
  • 00:17:50
    for physicians to get that because we
  • 00:17:52
    have been ingrained into clinical
  • 00:17:54
    evidence process in a different way
  • 00:17:56
    now and then the tools that we have been
  • 00:17:59
    using for years and years for clinical
  • 00:18:00
    research has been statistics
  • 00:18:03
    in statistics the whole goal is what
  • 00:18:05
    relationships to understand that if this
  • 00:18:08
    particular
  • 00:18:09
    situation correlates with this
  • 00:18:11
    particular disease smoking to cancer so
  • 00:18:14
    it doesn't matter it's still not
  • 00:18:16
    causality right we don't know if it's
  • 00:18:18
    truly causing right it's it's at the end
  • 00:18:19
    of the day it's a relationship that we
  • 00:18:21
    have identifying increased risk of
  • 00:18:22
    cancers and stroke and in smoking
  • 00:18:24
    patients and then we infer from that
  • 00:18:26
    particular statistics and that's why
  • 00:18:28
    it's very explainable because we know we
  • 00:18:30
    have run different kinds of
  • 00:18:33
    variations of statistics and then we
  • 00:18:34
    came to that particular conclusion
  • 00:18:37
    and the the artificial intelligence has
  • 00:18:39
    nothing to do with that artificial
  • 00:18:41
    intelligence is all about prediction
  • 00:18:43
    prediction prediction i'm sitting on a
  • 00:18:45
    card table and i want to know what the
  • 00:18:47
    next chord is so i can bet something
  • 00:18:50
    different that is it
  • 00:18:52
    the whole purpose of artificial
  • 00:18:54
    intelligence is to predict that's where
  • 00:18:56
    the real magic comes in if there's going
  • 00:18:58
    to be a financial decline or if there's
  • 00:19:01
    going to be this patient is going to
  • 00:19:02
    develop stroke in the next 10 years this
  • 00:19:05
    patient is going to develop heart attack
  • 00:19:06
    in the next five years we have lots and
  • 00:19:08
    lots of data we have millions of
  • 00:19:10
    examples and we have seen different
  • 00:19:12
    areas in which these patients perform
  • 00:19:14
    differently and now we can through a
  • 00:19:17
    significantly precise way predict
  • 00:19:21
    where that patient is going to fall in
  • 00:19:23
    the next five to 10 years and you can
  • 00:19:25
    change the trajectory that's the key
  • 00:19:27
    thing right we are in always responding
  • 00:19:31
    quite frankly most of the time in terms
  • 00:19:33
    of disease management and this is the
  • 00:19:36
    paradigm shift in terms of
  • 00:19:38
    disease management to moving from
  • 00:19:40
    treatment to prediction and prevention
  • 00:19:43
    so that's the key difference between
  • 00:19:45
    statistics and machine learning because
  • 00:19:47
    that's where we are going into in terms
  • 00:19:49
    of artificial intelligence and it is
  • 00:19:52
    truly based on human learning right i
  • 00:19:54
    mean if you look at the human learning i
  • 00:19:55
    mean if i look at one of the cat what
  • 00:19:57
    happens i have the one neuron that goes
  • 00:19:59
    into and then connects to different
  • 00:20:00
    synapses to my virtual uh visual cortex
  • 00:20:03
    and comes back and tells me okay this is
  • 00:20:05
    a cat and exactly in the same way
  • 00:20:08
    artificial neural networks are a design
  • 00:20:11
    it takes one of the images divides into
  • 00:20:13
    multiple areas and each single pixel is
  • 00:20:15
    sometimes
  • 00:20:17
    regulated in terms of what it is and
  • 00:20:19
    then eventually you get a complete
  • 00:20:21
    classification if this particular cat
  • 00:20:23
    scan or mri has an infarction tumor or
  • 00:20:26
    hemorrhage so what we're seeing is that
  • 00:20:28
    that it is that's why we call it
  • 00:20:30
    artificial intelligence because we are
  • 00:20:33
    trying to mimic human intelligence and
  • 00:20:35
    that's where the beauty of this whole
  • 00:20:37
    thing comes in
  • 00:20:38
    now
  • 00:20:39
    another thing you need to realize as far
  • 00:20:41
    as a difference is concerned is that
  • 00:20:44
    what is the difference between an
  • 00:20:45
    algorithm
  • 00:20:46
    and artificial intelligence look we
  • 00:20:49
    people don't understand they use the
  • 00:20:50
    word algorithm all the time and by the
  • 00:20:52
    way we have been using algorithms for 50
  • 00:20:54
    years this is a basic algorithm for
  • 00:20:56
    hyper neutrinia that we have used for
  • 00:20:59
    many many years right i mean i'm
  • 00:21:01
    assuming everyone has used that what's
  • 00:21:03
    this this is basically a rule-based
  • 00:21:05
    algorithm in which what you have done
  • 00:21:07
    what you have done is basically you have
  • 00:21:08
    sodium for this then you go true or
  • 00:21:10
    pseudo and then you go to next step then
  • 00:21:12
    you go to the next fork then you get to
  • 00:21:14
    the next fork order some studies and
  • 00:21:15
    then move forward this has been a
  • 00:21:17
    rule-based algorithm and as a matter of
  • 00:21:19
    fact to be honest with you medicine runs
  • 00:21:21
    on this rule-based algorithms if you
  • 00:21:23
    have a book with rule-based algorithms
  • 00:21:24
    you can give it to a first-year
  • 00:21:26
    medical student and he can get you
  • 00:21:28
    somewhere in terms of moving the patient
  • 00:21:30
    management forward
  • 00:21:32
    what's the difference then when you have
  • 00:21:34
    artificial intelligence
  • 00:21:36
    you see the rules are not made there are
  • 00:21:38
    no instructions
  • 00:21:40
    we give a million
  • 00:21:41
    records to the patient and tells them
  • 00:21:44
    well this is hyponatremia what do you
  • 00:21:45
    think it is as far as diagnosis is
  • 00:21:47
    concerned with the million examples it
  • 00:21:49
    finds out okay if the patient has these
  • 00:21:51
    characteristics then it's going to fall
  • 00:21:52
    into this bucket these characters fall
  • 00:21:54
    into this bucket these characters will
  • 00:21:56
    fall into this bucket creates its own
  • 00:21:58
    database algorithm and then when you
  • 00:22:00
    give another 10 000 new
  • 00:22:02
    testing protocol then it can predict
  • 00:22:05
    for you
  • 00:22:07
    where that patient is going to be
  • 00:22:08
    do you understand now i'm hoping that
  • 00:22:10
    this circles in because it's hard to put
  • 00:22:13
    these concepts into one sort of lecture
  • 00:22:16
    so this is the whole idea behind it who
  • 00:22:18
    will benefit from loop direct directs
  • 00:22:21
    and you're basically going to see okay
  • 00:22:22
    the data is going to basically tell you
  • 00:22:25
    who is going to do that rather than you
  • 00:22:27
    creating a rule-based system now
  • 00:22:30
    in the beginning there will be more
  • 00:22:31
    hybrid approaches right i mean i'm going
  • 00:22:32
    to basically use natural language
  • 00:22:34
    processing to
  • 00:22:36
    figure out all the data from the emr
  • 00:22:39
    then i'm going to have a then i'm going
  • 00:22:40
    to have part rule-based approach and
  • 00:22:42
    then i'm going to have something else so
  • 00:22:44
    there are different hybrid forms of care
  • 00:22:46
    pathways that are going on but
  • 00:22:47
    eventually it's going to evolve into one
  • 00:22:49
    scare pathway
  • 00:22:51
    so in algorithm versus ai data
  • 00:22:54
    codes
  • 00:22:55
    not rule codes now one of the other
  • 00:22:57
    things that are truly about artificial
  • 00:22:59
    intelligence is that if there's a
  • 00:23:01
    continuous stream of data those rules uh
  • 00:23:04
    that are effect the word rule with may
  • 00:23:06
    still be
  • 00:23:07
    the the way the artificial intelligence
  • 00:23:09
    predicts can change because there is a
  • 00:23:11
    continuous stream of data and as the
  • 00:23:12
    data changes the artificial intelligence
  • 00:23:14
    algorithm itself can change which is
  • 00:23:16
    very different from rule based because
  • 00:23:18
    rule based gets stuck right this is some
  • 00:23:20
    expert came in or whatever thousands of
  • 00:23:22
    different uh scientific observations
  • 00:23:24
    that created that rule-based algorithm
  • 00:23:26
    but the benefit of artificial
  • 00:23:28
    intelligence is that it can continuously
  • 00:23:30
    learn and that is the key thing however
  • 00:23:34
    that is not currently present in
  • 00:23:36
    medicine
  • 00:23:37
    let me give you an example
  • 00:23:39
    this art this continuous learning is
  • 00:23:41
    what youtube does facebook does twitter
  • 00:23:44
    does there's a constant new trends and
  • 00:23:46
    then they shift their algorithm models
  • 00:23:48
    accordingly and then prevent from
  • 00:23:50
    whatever reason and you can introduce
  • 00:23:52
    some of different ways in which to
  • 00:23:54
    introduce some bias and that's one of
  • 00:23:56
    the ways that they work but what we want
  • 00:23:58
    to do is that we should in medicine have
  • 00:24:00
    the same opportunity but unfortunately
  • 00:24:02
    that's not true
  • 00:24:04
    to date i am not aware of any medical
  • 00:24:07
    algorithm that is constantly learning
  • 00:24:10
    they have learned from the data that
  • 00:24:12
    algorithm has static is based on the
  • 00:24:15
    data and is applied but it is not
  • 00:24:18
    it is not constantly learning and the
  • 00:24:20
    reason behind it is that there's a
  • 00:24:22
    significant amount of clinical data
  • 00:24:24
    shift and clinician data shift and there
  • 00:24:26
    is their major issues um
  • 00:24:29
    let me give you an example because that
  • 00:24:31
    would be easier
  • 00:24:32
    look we have a fantastic algorithm for
  • 00:24:34
    sepsis for predicting sepsis and as a
  • 00:24:36
    matter of fact it worked beautifully
  • 00:24:38
    before covert but after covert the
  • 00:24:40
    sepsis was completely different because
  • 00:24:42
    most pre-covert sepsis was bacterially
  • 00:24:44
    origin but post
  • 00:24:46
    covid we had viral sepsis and those
  • 00:24:50
    viral substances were very different for
  • 00:24:52
    the computer to predict those viral
  • 00:24:54
    differences it needs new forms of data
  • 00:24:57
    so that it can predict much better we
  • 00:25:00
    didn't do that because it's not
  • 00:25:01
    consistently learning environment we
  • 00:25:03
    don't have it because of multiple
  • 00:25:04
    reasons but for now that's the biggest
  • 00:25:06
    issue because we do not have a
  • 00:25:07
    continuous learning environment that
  • 00:25:09
    sepsis algorithm freaking broke as a
  • 00:25:11
    matter of fact the company has to
  • 00:25:12
    retract it so that's the key issue in
  • 00:25:15
    terms of medicine and which is the the
  • 00:25:17
    biggest issue is the data data data high
  • 00:25:20
    quality data that you can process in
  • 00:25:22
    real time to produce predictions in real
  • 00:25:25
    time that could potentially be shifting
  • 00:25:27
    because of environment country
  • 00:25:29
    genetic background etc etc so that's the
  • 00:25:32
    big issue why we do not have continuous
  • 00:25:35
    learning of data engineers learning
  • 00:25:38
    algorithm which is the true form of
  • 00:25:39
    machine learning in medicine
  • 00:25:41
    so that's what another vision and that's
  • 00:25:43
    one of the key issues that another issue
  • 00:25:45
    is i keep asking me will ai will yeah i
  • 00:25:48
    replace physicians absolutely not
  • 00:25:50
    now it's already the current state if
  • 00:25:52
    you look at it it is basically combining
  • 00:25:54
    human and ai in terms of
  • 00:25:56
    very narrow tasks so if you have a
  • 00:25:58
    narrow task cd brain hemorrhage or no
  • 00:26:00
    hemorrhage yes but when i look at us cat
  • 00:26:03
    scan i mean it's not just like if it's a
  • 00:26:05
    bleed or not i mean it's a 70 80 year
  • 00:26:07
    old female that has significant atrophy
  • 00:26:09
    why the hell so i can talk about
  • 00:26:11
    dementia possibly with the same patient
  • 00:26:13
    while i'm talking about enzyme or
  • 00:26:14
    talking about stroke where i see that
  • 00:26:16
    the patient has calcification in places
  • 00:26:18
    where there were no calcifications so in
  • 00:26:20
    a complete narrow task artificial
  • 00:26:22
    intelligence is right there
  • 00:26:24
    in computer vision and i mean i'm going
  • 00:26:26
    to talk about they're definitely places
  • 00:26:28
    where artificial intelligence
  • 00:26:30
    significantly lacks and there are places
  • 00:26:32
    where artificial intelligence is better
  • 00:26:33
    than physicians
  • 00:26:35
    however in future we're going to see
  • 00:26:37
    that since we're going to have more and
  • 00:26:39
    more high quality data added we will are
  • 00:26:41
    building data pipelines in which
  • 00:26:43
    artificial intel data high quality data
  • 00:26:45
    is going to be input for continuous
  • 00:26:46
    learning we're going to see that data
  • 00:26:48
    the the process shift in which in more
  • 00:26:52
    and more tasks artificial intelligence
  • 00:26:54
    is going to get better and better from
  • 00:26:56
    humans i mean it's quite a while away
  • 00:26:58
    about 20 to 30 years away in terms of
  • 00:27:00
    truly seeing that massive impact in
  • 00:27:03
    medicine there are definitely areas in
  • 00:27:05
    which artificial intelligence is going
  • 00:27:07
    to be the standard
  • 00:27:09
    on which physician overlay reading will
  • 00:27:11
    be there and this is where we come in
  • 00:27:14
    e.i will not replace physicians however
  • 00:27:16
    physicians that use artificial
  • 00:27:18
    intelligence will replace a physician
  • 00:27:19
    who do not
  • 00:27:21
    that's something literally if this one
  • 00:27:23
    slide that you have to take care of is
  • 00:27:24
    this you have to learn it exactly as you
  • 00:27:27
    have learned everything else and
  • 00:27:28
    secondly initially at least in the
  • 00:27:30
    majority of cases you're going to see
  • 00:27:31
    that artificial intelligence and
  • 00:27:33
    automation that is added is basically
  • 00:27:36
    allowing you to have better decision
  • 00:27:39
    support in terms of faster care no
  • 00:27:41
    latency of care improved care outcomes
  • 00:27:43
    etc etc but at the end of the day you
  • 00:27:46
    have the helms of the
  • 00:27:49
    control in terms of patient management
  • 00:27:51
    is concerned think of it as the horse or
  • 00:27:52
    ski lag
  • 00:27:54
    physician
  • 00:27:55
    but the horse is getting better the
  • 00:27:56
    whole time ever i mean it literally it's
  • 00:27:59
    going from a horse to our ferrari so in
  • 00:28:01
    terms of moving in
  • 00:28:03
    in terms of improving accuracy and
  • 00:28:05
    improving predictions
  • 00:28:07
    so
  • 00:28:08
    now we come to development of artificial
  • 00:28:10
    intelligence as i talked about data and
  • 00:28:11
    data data look the data is is massively
  • 00:28:15
    improving as a matter of fact there's
  • 00:28:16
    one third of clinical data in unite in
  • 00:28:19
    production so if there's
  • 00:28:22
    100 data points are created about 33 of
  • 00:28:25
    them are coming from the health care
  • 00:28:26
    industry health care
  • 00:28:28
    is a massive
  • 00:28:30
    reason for expanding data in the world
  • 00:28:34
    and we are expanding data significantly
  • 00:28:36
    i mean youtube videos you know etc etc
  • 00:28:39
    so we are having a lot of data that is
  • 00:28:41
    coming out but healthcare still creates
  • 00:28:43
    a lot of data
  • 00:28:44
    and unfortunately that is
  • 00:28:45
    multi-dimensional data what i mean by
  • 00:28:47
    that is that there's disease data you
  • 00:28:49
    know radiomics genomics biomics and then
  • 00:28:52
    there's health data which we are you
  • 00:28:54
    know creating with our smart watches and
  • 00:28:56
    ecg machines voice analysis etc etc and
  • 00:28:59
    then there's claims data you know with
  • 00:29:00
    healthcare industries coming up with it
  • 00:29:02
    sorry healthcare insurance is coming in
  • 00:29:04
    and then of course we have clinical
  • 00:29:05
    trials data
  • 00:29:06
    bio data and clinical registries data
  • 00:29:08
    and those can be pulled on
  • 00:29:10
    which is important first of all there's
  • 00:29:12
    a lot of volume but unfortunately the
  • 00:29:15
    biggest curse in in
  • 00:29:17
    medicine is that it deman it's
  • 00:29:19
    dimensional over time as well so my body
  • 00:29:21
    metabolics is very different when i was
  • 00:29:23
    20 years old i mean i'm becoming an old
  • 00:29:25
    man now so clearly the metabolics have
  • 00:29:27
    changed so over time the data
  • 00:29:29
    characteristics change so that's where
  • 00:29:31
    the curse of dimensionality in
  • 00:29:32
    healthcare comes in
  • 00:29:34
    and then more importantly more and more
  • 00:29:36
    convergence is happening right i mean as
  • 00:29:37
    i said we have more digital therapeutics
  • 00:29:40
    are coming up which is going to create
  • 00:29:41
    new forms of data we're going to have
  • 00:29:43
    new forms of genetic analysis which
  • 00:29:45
    creates more data then we have new
  • 00:29:47
    manufacturing done so there was more
  • 00:29:48
    data so we're going to have more and
  • 00:29:50
    more data coming in into the pipeline
  • 00:29:52
    that no matter what you do you're going
  • 00:29:54
    to need artificial intelligence to churn
  • 00:29:56
    it period there's no other way around it
  • 00:29:58
    because it's just impossible for anyone
  • 00:30:00
    to take that into a data science and
  • 00:30:02
    then make sense of it because you have
  • 00:30:04
    to have that that data to be done and
  • 00:30:07
    that's where another thing comes in you
  • 00:30:09
    do have to have a purpose right you just
  • 00:30:11
    don't collect data for no reason you
  • 00:30:13
    collect data to process it that's key
  • 00:30:16
    thing that is why the development of
  • 00:30:18
    artificial intelligence is going to
  • 00:30:19
    happen in stacks so i don't know if you
  • 00:30:22
    have any brother or sister in in
  • 00:30:24
    software services but you should talk to
  • 00:30:26
    them they talk about software as a
  • 00:30:28
    service sas that's the key term software
  • 00:30:30
    as a services basically you find a small
  • 00:30:32
    problem and you develop an automation
  • 00:30:34
    solution on it and then you apply it to
  • 00:30:36
    the big stack and then you another
  • 00:30:37
    problem big stack another problem is
  • 00:30:39
    stack it's exactly the same artificial
  • 00:30:41
    intelligence is developing medicine i
  • 00:30:43
    have a problem in terms of artificial in
  • 00:30:45
    stroke i have medical imaging analysis
  • 00:30:48
    through computer vision one stack i need
  • 00:30:50
    natural language processing to process
  • 00:30:52
    patient reported data or also clinical
  • 00:30:55
    reports i have another uh
  • 00:30:58
    transfer and management issue that do
  • 00:30:59
    that i have another stack of genomics
  • 00:31:02
    all these stacks are currently going to
  • 00:31:03
    be in development on different stacks
  • 00:31:06
    and these are software services
  • 00:31:08
    eventually these software and services
  • 00:31:09
    is going to become immature and then
  • 00:31:11
    when they're going to become mature
  • 00:31:12
    they're going to combine and when
  • 00:31:14
    they're going to converge that's going
  • 00:31:15
    to turn into a full artificial
  • 00:31:17
    intelligence that is going to be more
  • 00:31:19
    more usable and believe me uh this is
  • 00:31:21
    disruptive and can happen suddenly like
  • 00:31:23
    the kodak moment right i mean nobody's
  • 00:31:25
    using plain films anymore everyone's
  • 00:31:27
    doing digital photography and i have
  • 00:31:29
    another lecture on the convergence of
  • 00:31:30
    virtual care and artificial intelligence
  • 00:31:32
    because the virtual care is now becoming
  • 00:31:34
    a very important part of producing
  • 00:31:36
    ground truth valuable confirmed
  • 00:31:39
    validated data in clinical so therefore
  • 00:31:42
    we're you know if you're interested in
  • 00:31:43
    looking at convergence more you can look
  • 00:31:45
    at that lecture
  • 00:31:46
    now
  • 00:31:48
    there are three different pillars of
  • 00:31:49
    artificial intelligence one of them is
  • 00:31:51
    computer vision which goes about digital
  • 00:31:53
    pathology etc then predictive analytics
  • 00:31:56
    and natural language processing
  • 00:31:58
    so let's go into that so computer vision
  • 00:32:00
    is all about neuroideology in your terms
  • 00:32:02
    and neuropathology and then in terms of
  • 00:32:04
    video analysis if the patient is walking
  • 00:32:05
    you can see if the patient has stranded
  • 00:32:08
    walk is it the spine issue is it the
  • 00:32:10
    multiple sclerosis patient in which we
  • 00:32:12
    are using image and video analysis to
  • 00:32:15
    quantify that walk and how the other
  • 00:32:17
    medic the new medication you're going to
  • 00:32:19
    develop is going to improve so we're not
  • 00:32:21
    going to talk about virtual clinical
  • 00:32:22
    trials right now but the idea is that
  • 00:32:24
    image and video analysis computer vision
  • 00:32:26
    then we have predictive analytics and
  • 00:32:28
    this is where the decision support and
  • 00:32:30
    trend analysis comes in simple pulse
  • 00:32:32
    oximetry is coming through the icu you
  • 00:32:33
    can say that if the pulse ox is changing
  • 00:32:35
    is the patient developing high increase
  • 00:32:37
    intracranial pressure or as the patient
  • 00:32:39
    has different heart rates and
  • 00:32:41
    blood pressure monitoring are changing
  • 00:32:43
    can you predict at least sometimes it as
  • 00:32:44
    a matter of fact their studies have
  • 00:32:46
    predicted a sudden uh development of
  • 00:32:48
    cardiac arrest in eyes sick sepsis
  • 00:32:51
    patients up to 15 minutes earlier so
  • 00:32:54
    that's where the prediction trend
  • 00:32:55
    analysis come in in the predictive
  • 00:32:56
    analytics analytics portion and then
  • 00:32:59
    finally natural language processing
  • 00:33:01
    natural language processing is the most
  • 00:33:02
    exciting and the most important feature
  • 00:33:05
    as a matter of fact um
  • 00:33:07
    that's why google shines a lot in terms
  • 00:33:10
    of you know its prowess in artificial
  • 00:33:12
    intelligence because that their whole
  • 00:33:14
    thing was behind search is natural
  • 00:33:17
    language processing all that word data
  • 00:33:19
    on the internet and how to collect
  • 00:33:21
    accumulate and create wisdom from it
  • 00:33:24
    there are two areas in it one is speech
  • 00:33:25
    and the other one is text speeches text
  • 00:33:27
    to speech speech to text all of them use
  • 00:33:29
    artificial intelligence and as far as
  • 00:33:31
    text is concerned classification of that
  • 00:33:34
    extraction of that summarization of that
  • 00:33:35
    translation of that as a matter of fact
  • 00:33:37
    ibm watson was
  • 00:33:39
    summarizing automatically
  • 00:33:41
    some of the cancer trials and while the
  • 00:33:43
    patient is being seen and when you
  • 00:33:45
    search on that they would recommend uh
  • 00:33:47
    which clinical trial is applicable to
  • 00:33:49
    that particular patient and pull up a
  • 00:33:51
    summary of that as well so clearly there
  • 00:33:53
    are different areas where there are
  • 00:33:55
    applications but this is a sort of a
  • 00:33:57
    chart for you to understand
  • 00:33:58
    what different areas of artificial
  • 00:34:00
    intelligence are and what they are
  • 00:34:01
    concentrating on
  • 00:34:03
    and we talked about data and convergence
  • 00:34:06
    data data data it is all about data
  • 00:34:08
    and uh data is literally the new oil
  • 00:34:11
    nobody talks about enron as the most
  • 00:34:13
    valuable company we talk about google
  • 00:34:14
    facebook tesla amazon and the reason
  • 00:34:17
    that tesla is the most valuable company
  • 00:34:18
    is because it doesn't have a freaking
  • 00:34:20
    four wheels that is better than any
  • 00:34:21
    other car because it is not a car
  • 00:34:23
    company it is a data company
  • 00:34:25
    and
  • 00:34:26
    one of the key things that you need to
  • 00:34:28
    keep working on and this is one of the
  • 00:34:30
    most important papers from last year is
  • 00:34:32
    that everyone wants to do the sexy model
  • 00:34:34
    work in artificial intelligence and i
  • 00:34:35
    want to create models and then turn
  • 00:34:37
    algorithms and everything
  • 00:34:39
    data work is the most important work if
  • 00:34:41
    you have a beautiful data like data
  • 00:34:43
    pipeline you will have
  • 00:34:46
    golden area to develop artificial
  • 00:34:48
    intelligence to develop ground truth and
  • 00:34:50
    once you have that you can be amazing in
  • 00:34:52
    terms of artificial so data work data
  • 00:34:54
    work data work it is not just modeling
  • 00:34:57
    and all that to be honest it's getting
  • 00:34:59
    very easy one of the other things you
  • 00:35:01
    have to remember whenever you provide
  • 00:35:03
    automation to anything automation is
  • 00:35:05
    like a catalyst if you have a process
  • 00:35:08
    that is garbage it's gonna catalyze
  • 00:35:10
    garbage into more garbage and if you
  • 00:35:12
    have an area in which you are efficient
  • 00:35:15
    and you put in a catalyst then that
  • 00:35:17
    efficiency is going to be expanded as
  • 00:35:18
    well so it's very important that
  • 00:35:21
    wherever artificial intelligence is
  • 00:35:22
    applied it is applied in a process that
  • 00:35:26
    has been optimized and efficient before
  • 00:35:28
    you apply artificial intelligence
  • 00:35:30
    because if you don't then that ai is
  • 00:35:32
    going to be a bigger problem in terms of
  • 00:35:35
    where it is because it basically is a
  • 00:35:37
    tool it is a catalyst anytime you have
  • 00:35:39
    that sort of automation make sure that
  • 00:35:41
    you have that
  • 00:35:42
    silk smooth out before you make
  • 00:35:44
    applications
  • 00:35:47
    now let's get to artificial intelligence
  • 00:35:49
    and neurology we're going to talk
  • 00:35:50
    quickly about two major things one is
  • 00:35:53
    stroke you see stroke is 861 billion
  • 00:35:56
    problem huge issue there's increasing
  • 00:35:58
    incidence increasing debts people are
  • 00:36:00
    living with it more more and more as a
  • 00:36:02
    matter of fact if you literally do one
  • 00:36:03
    minute delay you are basically cleaning
  • 00:36:06
    1.9 million neurons and 14 billion
  • 00:36:08
    synapses and if there's a fast
  • 00:36:10
    progressive stroke you can use a limb
  • 00:36:13
    lose a limb completely in five minutes
  • 00:36:16
    we definitely need to reduce time right
  • 00:36:18
    if you're not going to reduce time
  • 00:36:20
    we're going to have a big issue and
  • 00:36:22
    that's where artificial intelligence
  • 00:36:23
    comes in and this is one of the earliest
  • 00:36:26
    application of artificial intelligence
  • 00:36:27
    that is commercially available
  • 00:36:30
    and
  • 00:36:30
    viable
  • 00:36:31
    there's a difference commercially
  • 00:36:33
    available clearly this is fda approved
  • 00:36:34
    people have seen it people have applied
  • 00:36:36
    it i use this software every day in my
  • 00:36:38
    life and i'm providing stroke care and
  • 00:36:41
    it is getting reimbursed so people are
  • 00:36:43
    getting paid for the hard work and
  • 00:36:45
    artificial intelligence they have done
  • 00:36:46
    because it's just not about academics
  • 00:36:48
    it's about producing
  • 00:36:50
    a product that is commercially viable
  • 00:36:52
    and reimbursable
  • 00:36:54
    so
  • 00:36:54
    there are multiple software packages
  • 00:36:57
    this is one visa i don't know if you're
  • 00:36:58
    aware of dr adnan siddiqui and amir
  • 00:37:00
    hussain they are fantastic people here
  • 00:37:02
    in us amir is from egypt and atlanta
  • 00:37:04
    she's from pakistan who has done this
  • 00:37:06
    study and showed that artificial in
  • 00:37:08
    intelligence decreases latency to care
  • 00:37:10
    by 66 minutes think about the number of
  • 00:37:13
    limbs
  • 00:37:14
    you can save in lives you can save with
  • 00:37:16
    that because you can get to the patient
  • 00:37:18
    faster and earlier if you don't have a
  • 00:37:20
    thermectomy center it's to
  • 00:37:22
    apply it right you have an efficient
  • 00:37:24
    system
  • 00:37:25
    to get efficient results you just don't
  • 00:37:27
    put artificial intelligence where you
  • 00:37:29
    can't create a difference so remember
  • 00:37:31
    that but at the end of the day that
  • 00:37:33
    these are the value that artificial
  • 00:37:34
    intelligence is bringing we can see it
  • 00:37:36
    in interstable hemorrhage dissection we
  • 00:37:38
    can automate it into stable hemorrhagic
  • 00:37:40
    detection as we want we can have
  • 00:37:42
    significant amount of commercial
  • 00:37:43
    software that are available this is the
  • 00:37:45
    list
  • 00:37:46
    that is there and you have to find the
  • 00:37:48
    right problem to find the right solution
  • 00:37:51
    this is a
  • 00:37:53
    complete video that i did
  • 00:37:54
    make sure you look at it in terms of
  • 00:37:56
    role of artificial intelligence and
  • 00:37:57
    telestroke and how it
  • 00:38:00
    becomes more than life in terms of
  • 00:38:02
    decreasing the latency of care itself
  • 00:38:06
    and this is the paper that got published
  • 00:38:08
    about the same thing
  • 00:38:09
    in terms of role of artificial
  • 00:38:11
    intelligence and stroke and telestroke
  • 00:38:14
    the second big reason is biomarkers i
  • 00:38:16
    told you about it because one of the key
  • 00:38:18
    issues in
  • 00:38:19
    prediction of neurological disorders is
  • 00:38:22
    to have something that i can have to to
  • 00:38:25
    diagnose the patient and more
  • 00:38:27
    importantly not just the diagnosis early
  • 00:38:29
    diagnosis so i can make a change to
  • 00:38:31
    predict how the patient is going to move
  • 00:38:33
    and then any intervention early
  • 00:38:35
    intervention that can be done to correct
  • 00:38:38
    that bad pathological pathway that is
  • 00:38:40
    moving forward it is most important in
  • 00:38:43
    terms of stroke and in terms of
  • 00:38:47
    alzheimer's and everything and we're
  • 00:38:48
    working on it right it's not like we're
  • 00:38:50
    not stopped working on blood
  • 00:38:53
    based biomarkers we're looking at
  • 00:38:54
    hemorrhaging we're looking at
  • 00:38:56
    subarachnoid hemorrhage we're looking at
  • 00:38:57
    csf analysis in some of these patients
  • 00:38:59
    to find
  • 00:39:01
    lab-based biomarkers blood-based
  • 00:39:03
    serum-based biomarkers for these
  • 00:39:05
    patients however it is becoming harder
  • 00:39:07
    and harder and we have to produce
  • 00:39:09
    different methods of doing it and one of
  • 00:39:11
    the method is
  • 00:39:13
    using deep learning artificial
  • 00:39:15
    interlearning in a multimodal way to
  • 00:39:18
    predict
  • 00:39:19
    early if there is anything going on and
  • 00:39:21
    one of the best applications this has
  • 00:39:23
    been in alzheimer's in which what we're
  • 00:39:25
    seeing is that we can combine imaging
  • 00:39:27
    genetics and clinical test data this is
  • 00:39:30
    not available commercially this is
  • 00:39:32
    currently in research only however what
  • 00:39:36
    we are seeing the trend going forward is
  • 00:39:38
    that when we have combination of these
  • 00:39:41
    different data sources i can predict
  • 00:39:44
    alzheimer's at the age of 55 what am i
  • 00:39:46
    going to do with it well currently
  • 00:39:47
    nothing i mean there's only one drug
  • 00:39:49
    that is available which is highly
  • 00:39:50
    controversial but at the end of the day
  • 00:39:52
    if i have to develop a drug i cannot
  • 00:39:54
    give it to a person who already has
  • 00:39:56
    advanced dementia what am i going to do
  • 00:39:57
    the brain is already
  • 00:39:59
    problematic we need to early intervene
  • 00:40:01
    so the minute i'm able to have
  • 00:40:03
    authenticated clinically validated
  • 00:40:06
    digital or multi-modal biomarker i can
  • 00:40:09
    intervene with a new medication study
  • 00:40:11
    that medication and see how it goes so
  • 00:40:14
    we're going to see that more and more as
  • 00:40:16
    far as biomarkers are concerned
  • 00:40:18
    now i have all of this and more in terms
  • 00:40:21
    of my courses that are coming
  • 00:40:23
    as far as how to choose a medical
  • 00:40:25
    specialty in this new post
  • 00:40:27
    pandemic digital healthcare era in which
  • 00:40:29
    i'm going to talk about artificial
  • 00:40:30
    intelligence virtual care and advanced
  • 00:40:32
    practice providers and i'm going to be
  • 00:40:35
    giving out a total of 100 participants
  • 00:40:37
    significant discount in honor of dr
  • 00:40:40
    wasse so i'm going to send it because he
  • 00:40:42
    is a beacon of neurology in pakistan and
  • 00:40:45
    i have been impressed that how he has
  • 00:40:48
    managed to do all this all the time so
  • 00:40:50
    i'm going to send it to him and he can
  • 00:40:52
    distribute the codes to anyone and
  • 00:40:54
    people can take this other course in
  • 00:40:57
    much more detail as far as artificial
  • 00:40:59
    intelligence virtual care and other
  • 00:41:01
    things like lifestyle considerations
  • 00:41:02
    financial considerations etc so it's
  • 00:41:05
    being built it's going to be available
  • 00:41:06
    next month and i hope to see you there
  • 00:41:08
    as well
  • 00:41:09
    another thing that you might want to
  • 00:41:10
    help me with is about
  • 00:41:12
    neurologypocketbook.com
  • 00:41:14
    you can go there multiple chapters that
  • 00:41:15
    i'm adding about digital neurology in
  • 00:41:17
    terms of what is digital health what is
  • 00:41:19
    m health
  • 00:41:21
    mobile health in neurology and then
  • 00:41:23
    there are chapters on digital education
  • 00:41:25
    in neurology artificial intelligence
  • 00:41:27
    neurology like staple ischemia and this
  • 00:41:28
    is all open access so you can copy paste
  • 00:41:31
    it put it in your presentations etc and
  • 00:41:33
    maybe contribute if you want to
  • 00:41:35
    this is the website and then of course
  • 00:41:37
    everything that i do is on one link tree
  • 00:41:39
    including twitter etc all that thank you
  • 00:41:43
    so much for your help and i'm very very
  • 00:41:45
    grateful for your time if you save a
  • 00:41:46
    life it is as if you saved the life of
  • 00:41:48
    mankind please make sure you like
  • 00:41:50
    comment share and subscribe to my
  • 00:41:52
    youtube channel and my newsletter if you
  • 00:41:55
    want to get in touch with me the best
  • 00:41:56
    way is to go through twitter or while
  • 00:41:58
    linkedin also make sure you follow the
  • 00:42:00
    academy website for regular updates
  • 00:42:02
    thank you so much
الوسوم
  • Artificial Intelligence
  • Neurology
  • Stroke
  • Epilepsy
  • Healthcare
  • Machine Learning
  • Deep Learning
  • Biomarkers
  • Patient Care
  • Medical Technology