1
00:00:00,667 --> 00:00:02,770
So, well, I do applied math,

2
00:00:02,770 --> 00:00:04,294
and this is a peculiar problem

3
00:00:04,294 --> 00:00:06,467
for anyone who does applied math, is that

4
00:00:06,467 --> 00:00:08,400
we are like management consultants.

5
00:00:08,400 --> 00:00:10,346
No one knows what the hell we do.

6
00:00:10,346 --> 00:00:12,620
So I am going to give you some -- attempt today

7
00:00:12,620 --> 00:00:14,913
to try and explain to you what I do.

8
00:00:14,913 --> 00:00:18,234
So, dancing is one of the most human of activities.

9
00:00:18,234 --> 00:00:21,916
We delight at ballet virtuosos and tap dancers

10
00:00:21,916 --> 00:00:23,064
you will see later on.

11
00:00:23,064 --> 00:00:25,754
Now, ballet requires an extraordinary level of expertise

12
00:00:25,754 --> 00:00:28,668
and a high level of skill,

13
00:00:28,668 --> 00:00:31,199
and probably a level of initial suitability

14
00:00:31,199 --> 00:00:33,046
that may well have a genetic component to it.

15
00:00:33,046 --> 00:00:36,439
Now, sadly, neurological disorders such as Parkinson's disease

16
00:00:36,439 --> 00:00:38,526
gradually destroy this extraordinary ability,

17
00:00:38,526 --> 00:00:40,849
as it is doing to my friend Jan Stripling, who was

18
00:00:40,849 --> 00:00:43,816
a virtuoso ballet dancer in his time.

19
00:00:43,816 --> 00:00:46,870
So great progress and treatment has been made over the years.

20
00:00:46,870 --> 00:00:49,814
However, there are 6.3 million people worldwide

21
00:00:49,814 --> 00:00:53,262
who have the disease, and they have to live with

22
00:00:53,262 --> 00:00:55,830
incurable weakness, tremor, rigidity

23
00:00:55,830 --> 00:00:57,687
and the other symptoms that go along with the disease,

24
00:00:57,687 --> 00:01:00,070
so what we need are objective tools

25
00:01:00,070 --> 00:01:03,127
to detect the disease before it's too late.

26
00:01:03,127 --> 00:01:05,681
We need to be able to measure progression objectively,

27
00:01:05,681 --> 00:01:08,854
and ultimately, the only way we're going to know

28
00:01:08,854 --> 00:01:11,046
when we actually have a cure is when we have

29
00:01:11,046 --> 00:01:14,444
an objective measure that can answer that for sure.

30
00:01:14,444 --> 00:01:17,294
But frustratingly, with Parkinson's disease

31
00:01:17,294 --> 00:01:19,647
and other movement disorders, there are no biomarkers,

32
00:01:19,647 --> 00:01:21,876
so there's no simple blood test that you can do,

33
00:01:21,876 --> 00:01:23,678
and the best that we have is like

34
00:01:23,678 --> 00:01:25,919
this 20-minute neurologist test.

35
00:01:25,919 --> 00:01:28,377
You have to go to the clinic to do it. It's very, very costly,

36
00:01:28,377 --> 00:01:31,134
and that means that, outside the clinical trials,

37
00:01:31,134 --> 00:01:33,862
it's just never done. It's never done.

38
00:01:33,862 --> 00:01:36,939
But what if patients could do this test at home?

39
00:01:36,939 --> 00:01:39,037
Now, that would actually save on a difficult trip to the clinic,

40
00:01:39,037 --> 00:01:43,291
and what if patients could do that test themselves, right?

41
00:01:43,291 --> 00:01:45,211
No expensive staff time required.

42
00:01:45,211 --> 00:01:46,629
Takes about $300, by the way,

43
00:01:46,629 --> 00:01:48,622
in the neurologist's clinic to do it.

44
00:01:48,622 --> 00:01:51,303
So what I want to propose to you as an unconventional way

45
00:01:51,303 --> 00:01:52,817
in which we can try to achieve this,

46
00:01:52,817 --> 00:01:54,625
because, you see, in one sense, at least,

47
00:01:54,625 --> 00:01:57,881
we are all virtuosos like my friend Jan Stripling.

48
00:01:57,881 --> 00:02:01,636
So here we have a video of the vibrating vocal folds.

49
00:02:01,636 --> 00:02:04,865
Now, this is healthy and this is somebody making speech sounds,

50
00:02:04,865 --> 00:02:08,329
and we can think of ourselves as vocal ballet dancers,

51
00:02:08,329 --> 00:02:10,543
because we have to coordinate all of these vocal organs

52
00:02:10,543 --> 00:02:12,838
when we make sounds, and we all actually

53
00:02:12,838 --> 00:02:15,134
have the genes for it. FoxP2, for example.

54
00:02:15,134 --> 00:02:17,847
And like ballet, it takes an extraordinary level of training.

55
00:02:17,847 --> 00:02:20,432
I mean, just think how long it takes a child to learn to speak.

56
00:02:20,432 --> 00:02:22,814
From the sound, we can actually track

57
00:02:22,814 --> 00:02:25,095
the vocal fold position as it vibrates,

58
00:02:25,095 --> 00:02:27,638
and just as the limbs are affected in Parkinson's,

59
00:02:27,638 --> 00:02:30,419
so too are the vocal organs.

60
00:02:30,419 --> 00:02:32,299
So on the bottom trace, you can see an example of

61
00:02:32,299 --> 00:02:33,997
irregular vocal fold tremor.

62
00:02:33,997 --> 00:02:35,165
We see all the same symptoms.

63
00:02:35,165 --> 00:02:38,095
We see vocal tremor, weakness and rigidity.

64
00:02:38,095 --> 00:02:40,199
The speech actually becomes quieter and more breathy

65
00:02:40,199 --> 00:02:42,432
after a while, and that's one of the example symptoms of it.

66
00:02:42,432 --> 00:02:45,279
So these vocal effects can actually be quite subtle,

67
00:02:45,279 --> 00:02:48,495
in some cases, but with any digital microphone,

68
00:02:48,495 --> 00:02:51,040
and using precision voice analysis software

69
00:02:51,040 --> 00:02:53,449
in combination with the latest in machine learning,

70
00:02:53,449 --> 00:02:55,027
which is very advanced by now,

71
00:02:55,027 --> 00:02:57,913
we can now quantify exactly where somebody lies

72
00:02:57,913 --> 00:03:00,794
on a continuum between health and disease

73
00:03:00,794 --> 00:03:03,390
using voice signals alone.

74
00:03:03,390 --> 00:03:05,704
So these voice-based tests, how do they stack up against

75
00:03:05,704 --> 00:03:07,854
expert clinical tests? We'll, they're both non-invasive.

76
00:03:07,854 --> 00:03:11,836
The neurologist's test is non-invasive. They both use existing infrastructure.

77
00:03:11,836 --> 00:03:14,840
You don't have to design a whole new set of hospitals to do it.

78
00:03:14,840 --> 00:03:17,142
And they're both accurate. Okay, but in addition,

79
00:03:17,142 --> 00:03:20,469
voice-based tests are non-expert.

80
00:03:20,469 --> 00:03:22,461
That means they can be self-administered.

81
00:03:22,461 --> 00:03:25,041
They're high-speed, take about 30 seconds at most.

82
00:03:25,041 --> 00:03:27,335
They're ultra-low cost, and we all know what happens.

83
00:03:27,335 --> 00:03:29,775
When something becomes ultra-low cost,

84
00:03:29,775 --> 00:03:32,071
it becomes massively scalable.

85
00:03:32,071 --> 00:03:35,746
So here are some amazing goals that I think we can deal with now.

86
00:03:35,746 --> 00:03:38,172
We can reduce logistical difficulties with patients.

87
00:03:38,172 --> 00:03:40,484
No need to go to the clinic for a routine checkup.

88
00:03:40,484 --> 00:03:42,804
We can do high-frequency monitoring to get objective data.

89
00:03:42,804 --> 00:03:46,909
We can perform low-cost mass recruitment for clinical trials,

90
00:03:46,909 --> 00:03:49,024
and we can make population-scale screening

91
00:03:49,024 --> 00:03:50,620
feasible for the first time.

92
00:03:50,620 --> 00:03:52,822
We have the opportunity to start to search

93
00:03:52,822 --> 00:03:56,363
for the early biomarkers of the disease before it's too late.

94
00:03:56,363 --> 00:03:59,121
So, taking the first steps towards this today,

95
00:03:59,121 --> 00:04:01,247
we're launching the Parkinson's Voice Initiative.

96
00:04:01,247 --> 00:04:03,479
With Aculab and PatientsLikeMe, we're aiming

97
00:04:03,479 --> 00:04:05,407
to record a very large number of voices worldwide

98
00:04:05,407 --> 00:04:08,547
to collect enough data to start to tackle these four goals.

99
00:04:08,547 --> 00:04:10,247
We have local numbers accessible to three quarters

100
00:04:10,247 --> 00:04:11,857
of a billion people on the planet.

101
00:04:11,857 --> 00:04:14,934
Anyone healthy or with Parkinson's can call in, cheaply,

102
00:04:14,934 --> 00:04:17,073
and leave recordings, a few cents each,

103
00:04:17,073 --> 00:04:19,263
and I'm really happy to announce that we've already hit

104
00:04:19,263 --> 00:04:22,806
six percent of our target just in eight hours.

105
00:04:22,806 --> 00:04:26,557
Thank you. (Applause)

106
00:04:26,557 --> 00:04:32,877
(Applause)

107
00:04:32,877 --> 00:04:36,452
Tom Rielly: So Max, by taking all these samples of,

108
00:04:36,452 --> 00:04:39,228
let's say, 10,000 people,

109
00:04:39,228 --> 00:04:42,082
you'll be able to tell who's healthy and who's not?

110
00:04:42,082 --> 00:04:43,767
What are you going to get out of those samples?

111
00:04:43,767 --> 00:04:45,597
Max Little: Yeah. Yeah. So what will happen is that,

112
00:04:45,597 --> 00:04:47,254
during the call you have to indicate whether or not

113
00:04:47,254 --> 00:04:48,521
you have the disease or not, you see. TR: Right.

114
00:04:48,521 --> 00:04:51,028
ML: You see, some people may not do it. They may not get through it.

115
00:04:51,028 --> 00:04:53,745
But we'll get a very large sample of data that is collected

116
00:04:53,745 --> 00:04:57,153
from all different circumstances, and it's getting it

117
00:04:57,153 --> 00:04:59,058
in different circumstances that matter because then

118
00:04:59,058 --> 00:05:02,442
we are looking at ironing out the confounding factors,

119
00:05:02,442 --> 00:05:04,603
and looking for the actual markers of the disease.

120
00:05:04,603 --> 00:05:07,100
TR: So you're 86 percent accurate right now?

121
00:05:07,100 --> 00:05:08,294
ML: It's much better than that.

122
00:05:08,294 --> 00:05:10,014
Actually, my student Thanasis, I have to plug him,

123
00:05:10,014 --> 00:05:11,884
because he's done some fantastic work,

124
00:05:11,884 --> 00:05:15,654
and now he has proved that it works over the mobile telephone network as well,

125
00:05:15,654 --> 00:05:19,044
which enables this project, and we're getting 99 percent accuracy.

126
00:05:19,044 --> 00:05:20,620
TR: Ninety-nine. Well, that's an improvement.

127
00:05:20,620 --> 00:05:22,821
So what that means is that people will be able to —

128
00:05:22,821 --> 00:05:24,673
ML: (Laughs)

129
00:05:24,673 --> 00:05:26,579
TR: People will be able to call in from their mobile phones

130
00:05:26,579 --> 00:05:29,651
and do this test, and people with Parkinson's could call in,

131
00:05:29,651 --> 00:05:32,521
record their voice, and then their doctor can check up

132
00:05:32,521 --> 00:05:35,202
on their progress, see where they're doing in this course of the disease.

133
00:05:35,202 --> 00:05:36,172
ML: Absolutely.

134
00:05:36,172 --> 00:05:37,915
TR: Thanks so much. Max Little, everybody.

135
00:05:37,915 --> 00:05:43,072
ML: Thanks, Tom. (Applause)