1
00:00:00,930 --> 00:00:03,347
(slow music)

2
00:00:09,570 --> 00:00:12,790
- It is a leading cause
of disability worldwide.

3
00:00:12,790 --> 00:00:15,530
And it's defined by the persistent feeling

4
00:00:15,530 --> 00:00:17,570
of negative emotions.

5
00:00:17,570 --> 00:00:19,053
I'm talking about depression.

6
00:00:19,970 --> 00:00:23,730
Imagine feeling helpless,
worthless, exhausted,

7
00:00:23,730 --> 00:00:25,720
not even knowing why.

8
00:00:25,720 --> 00:00:26,830
It is treatable,

9
00:00:26,830 --> 00:00:30,490
but most cases go by
unnoticed or misdiagnosed,

10
00:00:30,490 --> 00:00:33,410
severely impacting your
relationships, work,

11
00:00:33,410 --> 00:00:36,300
and the ability to
function as a normal human.

12
00:00:36,300 --> 00:00:37,770
And in worst cases,

13
00:00:37,770 --> 00:00:40,600
it can slowly push the
person away towards suicide

14
00:00:40,600 --> 00:00:42,970
without their loved one realising,

15
00:00:42,970 --> 00:00:46,906
such as the case of Chester
Bennington and Robin Williams.

16
00:00:46,906 --> 00:00:50,474
Depression is caused by a
combination of psychological,

17
00:00:50,474 --> 00:00:53,470
biological, and social sources of distress

18
00:00:53,470 --> 00:00:56,030
that alter neural circuits in your brain,

19
00:00:56,030 --> 00:00:59,060
leading to behavioural
and physical symptoms.

20
00:00:59,060 --> 00:01:02,580
Those symptoms are translated
into our outward appearance.

21
00:01:02,580 --> 00:01:04,780
Like most of our facial expressions,

22
00:01:04,780 --> 00:01:09,250
like how often we smile,
blink, or give eye contact.

23
00:01:09,250 --> 00:01:12,580
My research investigates the
use of artificial intelligence

24
00:01:12,580 --> 00:01:15,770
to detect symptoms of
depression from one's face

25
00:01:15,770 --> 00:01:18,720
to help doctors make a
more informed decision.

26
00:01:18,720 --> 00:01:22,600
Artificial intelligence and
machine learning from examples

27
00:01:22,600 --> 00:01:25,514
and its own experiences
to make future decisions.

28
00:01:25,514 --> 00:01:28,630
It recognises patterns and how they relate

29
00:01:28,630 --> 00:01:31,170
by looking at large amounts of data.

30
00:01:31,170 --> 00:01:33,620
But can it recognise
patterns of depression

31
00:01:33,620 --> 00:01:35,530
by just looking at us?

32
00:01:35,530 --> 00:01:37,680
To check, I've prepared video footage

33
00:01:37,680 --> 00:01:40,460
of around 120 participants,

34
00:01:40,460 --> 00:01:42,500
each labelled by a psychologist

35
00:01:42,500 --> 00:01:44,673
with a severity of depression.

36
00:01:46,230 --> 00:01:49,420
Using an algorithm in
AI, I trained a model

37
00:01:49,420 --> 00:01:51,890
to learn symptoms of depression.

38
00:01:51,890 --> 00:01:54,420
The model learns by
breaking down video frames

39
00:01:54,420 --> 00:01:55,560
into smaller sections,

40
00:01:55,560 --> 00:01:58,400
such as areas of the mouth and the eyes.

41
00:01:58,400 --> 00:02:01,950
It then learns characteristics
within these areas,

42
00:02:01,950 --> 00:02:03,330
links it back to the face,

43
00:02:03,330 --> 00:02:06,250
and finally, it's depression severity.

44
00:02:06,250 --> 00:02:08,060
By just looking at faces,

45
00:02:08,060 --> 00:02:11,460
the model was able to predict
severities of depression

46
00:02:11,460 --> 00:02:14,660
with an accuracy that's
way better than chance.

47
00:02:14,660 --> 00:02:17,350
In the future, I want to
investigate more techniques

48
00:02:17,350 --> 00:02:21,480
used in AI to help
improve model performance.

49
00:02:21,480 --> 00:02:24,440
My experiment is a step
closer to utilising

50
00:02:24,440 --> 00:02:27,670
the most advanced techniques
to help our doctors

51
00:02:27,670 --> 00:02:32,420
with a future having less cases
untreated or misdiagnosed.

52
00:02:32,420 --> 00:02:34,390
Imagine your doctor routinely checking

53
00:02:34,390 --> 00:02:36,520
for any symptoms of
depression during a visit,

54
00:02:36,520 --> 00:02:39,230
just like you get your blood
pressure and weight checked.

55
00:02:39,230 --> 00:02:42,360
Even better so, my technique
can be used remotely

56
00:02:42,360 --> 00:02:44,760
at the convenience of your home.

57
00:02:44,760 --> 00:02:47,600
It is time to save
ourselves and loved ones

58
00:02:47,600 --> 00:02:49,110
from slipping away.

59
00:02:49,110 --> 00:02:50,950
By looking through the lens of AI,

60
00:02:50,950 --> 00:02:53,360
maybe we can understand.

61
00:02:53,360 --> 00:02:54,193
Thank you.

62
00:02:55,066 --> 00:02:57,483
(slow music)