How AI can make health care better

0:01the world is facing a big medical

0:03problem

0:04a growing number of patients

0:07and not enough doctors to treat them

0:10so could artificial intelligence be the

0:12cure i feel like i'm working at the

0:14forefront of something that could

0:16potentially be revolutionary to

0:18healthcare in the future ai has the

0:20power to transform the ways patients are

0:22diagnosed and treated when we think

0:25it'll become a game changer and to make

0:27the testing of new medical procedures

0:30more efficient and effective if we can

0:32get devices that can be developed faster

0:36better

0:37and quicker

0:38then there are huge benefits

0:42[Music]

0:56elaine manor is blind in one eye

1:01she is a victim of age-related macular

1:04degeneration

1:06the most common cause of blindness in

1:08the uk and us

1:11when it went to my other eye i was just

1:14terrified i was in bits i was weeping in

1:18the rain and

1:19thinking i don't want to be here

1:24[Music]

1:30for years the threat of losing her other

1:32eye has loomed large

1:34until a successful treatment enabled the

1:37thankful 75 year old to see her way to

1:40some high wire fundraising

1:42i did the high sick wire in europe

1:47i did the sky dive

1:49and then i did the the wing walk

1:53[Music]

1:58but the doctor who saved elaine's eyes

2:00says she is one of the lucky ones nearly

2:0310 of all clinic appointments in the nhs

2:05are for eyes that's nearly 10 million

2:08appointments per year so to put it

2:11brutally we're almost drowning in the

2:13number of patients we need to see

2:15and as a result of that there are some

2:16patients unfortunately who go blind

2:18because of delays in being seen and

2:20treated a pregnant mother was left

2:23almost completely blind waiting for care

2:26she has since given birth and has never

2:29seen her daughter's

2:31face dr keen

2:33believes there's an answer

2:35artificial intelligence

2:38he and his partners have developed ais

2:41which can diagnose over 50 types of eye

2:44disease just as well as a doctor

2:47but do so

2:48much much more quickly the ai system can

2:51analyze the retinal scans within seconds

2:54and it can delineate all of the

2:56different disease features on the scans

2:58a human expert would probably take hours

3:00or even days to complete the same task

3:03ai can help to address a growing global

3:06challenge

3:07in 2020

3:09an estimated 596

3:12million people had distance vision

3:14impairment worldwide

3:16of whom 43 million were blind

3:19by 2050 both these figures are set to

3:23increase by approximately 50 percent

3:26a task that can take specialist doctors

3:28hours now being done in seconds through

3:31artificial intelligence and it's not

3:33just eyesight ai's ability to mine and

3:37analyze patient data far more quickly

3:39than humans can mean diagnoses could

3:42improve in many areas of medicine we do

3:44more than a thousand scans per day in

3:47the hospital it's a challenge because

3:50where do we get the the human experts to

3:52be able to review all those scans but

3:53it's also an opportunity because that

3:55huge amount of data is the perfect

3:57substrate for the development of

3:59artificial intelligence systems

4:02but there are concerns

4:04in particular threats to the privacy of

4:07patients

4:08deepmind google's ai company and one of

4:12dr keane's partners has found itself

4:14under fire

4:16google deep mind the search giant's

4:18artificial intelligence arm may have

4:20received the personally identifying

4:22medical records of 1.6 million nhs

4:24patients here at the royal fair hospital

4:27on a legally inappropriate basis

4:30unrelated to its work with dr keane

4:32deepmind is currently facing legal

4:34action over its use of nhs data

4:38yet if data can be better protected

4:40ai has the capacity to make patient care

4:43much better

4:44and more efficient

4:46so we have a world that is essentially

4:49very much connected

4:50but yet healthcare data is siloed

4:53we can order a taxi from almost anywhere

4:56in the world using our smartphones

4:58but yet if we have a patient who comes

5:00to an eye hospital like moorefields but

5:02they're also attending a hospital

5:05because they've got cancer we often

5:06can't easily connect their data

5:12dr keen hopes his latest collaboration

5:14with machine learning startup bitfont

5:17could not only join data dots better but

5:20also improve patient privacy with fount

5:23is is a kind of switchboard all we do is

5:26essentially pass messages between

5:28someone who wants to ask something of

5:30the data set and the owner of the data

5:33the data

5:34never never leaves its home location so

5:37if that data is held by a hospital

5:39no data ever leaves the hospital bitfont

5:41says this technology could have other

5:43benefits like approving new treatments

5:46more quickly and safely patients are

5:49losing out a lot by the fact that ideal

5:51medical treatments for them are not

5:53coming through to to market

5:55with the extra technical guarantees that

5:58privacy preserving techniques like

6:00bitfont can provide there's been a real

6:03feeling around the healthcare ecosystem

6:06that that could speed up a lot of those

6:08governance processes

6:12by 2027

6:14ai's value in the healthcare market is

6:16expected to be eight times bigger than

6:18in 2020 growth could also be boosted if

6:22clinicians reduce their dependence on

6:24coders and start to develop their own ai

6:27systems

6:28this is really exciting because today's

6:30retinal experts have been unable to

6:32identify gender dr kira o'bern is part

6:36of a team of clinicians who have managed

6:37to do what google brain did in 2018

6:41develop ai that can recognize gender

6:43from retinal scans

6:45something no human can do

6:47a member of our research group developed

6:49a code-free deep learning model which

6:51accurately identified gender from

6:53rational images

6:54this is incredibly exciting because by

6:56positioning clinicians

6:58to develop their own models

7:00independently it could really open the

7:02door to further discoveries in both

7:04disease patterns and disease biomarkers

7:07a new generation of doctors believe

7:09empowering clinicians in this way will

7:12bring them closer to patients

7:14generally it's the clinician that's the

7:16healthcare workers working face-to-face

7:18with the patient to understand what the

7:21patients need best

7:23so therefore i believe that if we can

7:25allow them to independently develop

7:27their own tools this will allow the

7:29patient to remain at the very forefront

7:31of everything

7:33it's the world's first hand-held

7:35battery-powered computer able to hold

7:37thousands of data points i think that

7:40we're potentially at a tipping point a

7:42little bit like the tipping point that

7:44we saw in the late 1970s in the

7:46computing industry we had the

7:48introduction of the first personal

7:49computers if you empower people with

7:52this technology even if primitive in the

7:54beginning they will come up with

7:56hundreds or even thousands of

7:57applications that the engineers would

7:59never have thought of

8:03but some are skeptical about pinning

8:05hopes too fast on ai

8:08the issue is that ai models are

8:09essentially black boxes and so what

8:12happens is that when they're working

8:13well they're working well and no

8:15questions are asked but what happens if

8:17a wrong decision is made what happens

8:20when something goes wrong and how do we

8:21really trace that back and ensure

8:23accountability and guarantee

8:26interpretability if we're using these

8:27black box models

8:30while it is early days for ai in

8:32medicine it could also improve the

8:34testing of new medical devices

8:36a long time ago

8:38a million years bc

8:41everything was

8:43absolutely free

8:46in 2021 former part-time singer and

8:49model patricia walker had an artificial

8:51valve inserted into her heart to save

8:54her life i needed a new valve

8:56because the one that i had wasn't

8:58working it was dripping this is why i

9:00was feeling

9:02the pain that i was getting the

9:03exhaustion and it it was getting

9:07worse

9:08although patricia's operation was a

9:10success

9:11it was not without risk and the

9:13cardiologist who inserted the valve says

9:16ai can make new technologies like this

9:19safer for patients

9:21by creating virtual trials

9:24if we can plan a procedure by simulating

9:27that procedure in an individual patient

9:30in a computer-generated model before we

9:32go to the patient then we can get a much

9:34better outcome in the patient without

9:35posing any risk to that patient at all

9:39at the university of leeds dr blackman

9:42is collaborating with professor alex

9:44frangie he is using machine learning to

9:47automate the building of

9:48three-dimensional digital replicas this

9:50is the real one

9:52and this is the fake one

9:55but it's very difficult to the naked eye

9:56to to pick that up

9:58virtual trials mean multiple variations

10:00of a proposed new procedure or

10:03technology can be tried out we can test

10:06different scenarios of treatment on the

10:08same

10:09anatomy and physiology of a given

10:12virtual individual and that's something

10:14which is not again possible to do with

10:15conventional trials

10:18we're also comparing different scenarios

10:20of hypertensive and normal tensive

10:22conditions

10:23virtual trials do not replace human

10:26trials

10:27but they do speed up the time and reduce

10:29the money required to identify the right

10:31devices and humans to test

10:35in 2021 the team at the university of

10:38leeds found their virtual trials

10:40produced the same results as three

10:42clinical trials but much more

10:44efficiently

10:46each of those studies took between six

10:48and eight years to to be undertaken and

10:51they probably cost around 20 to 30

10:53million each of them so what we showed

10:56is that in this computational study that

10:59the execution of it took about three

11:00months so

11:02that's you know less than

11:0520 grand

11:06some working in the medical world

11:08believe a bright future lies ahead as

11:11ais become more sophisticated and more

11:14capable

11:15in other words more intelligent ai 1.0

11:19in my view is the ability to automate

11:22tasks that otherwise

11:24could be very boring or time consuming

11:26or repetitive

11:29ai 2.0

11:31is the one that actually tries to look

11:33at incorporating prior information on

11:35the physics on the physiology in a much

11:38more intimate manner with the data so

11:40it's not just data driven but it's also

11:42knowledge driven for healthcare it's

11:44hard to see a future without ai

11:47i strongly believe that one day

11:51artificial intelligence

11:54will renew this eye

12:04thanks for watching i'm tom standage

12:06deputy editor of the economist to read

12:09more of our coverage of ai please click

12:11on the link don't forget to subscribe

12:27you