Digital Health für Einsteiger

1. 00:00Welcome to the second part of our Connected Healthcare series.
2. 00:04In the first part, I had given an overview to give an idea, what Connected Healthcare actually is.
3. 00:14I would like to recap with a slide.
4. 00:19So, we're mainly about, Building personal electronic health assistants,
5. 00:26That continuously measure health-related data from our everyday lives, detect certain health conditions and then suggest how we can lead a healthier lifestyle, for example.
6. 00:42And from that definition we get the three important areas that we work on.
7. 00:48So if we want to collect something continuously, we need sensors, that work continuously and that, above all, are comfortable,
8. 00:58So that, in the best case, are built into our clothes, our watches, in our environment, so that we don't even notice that they are there.
9. 01:06These sensors then provide us with a lot of data, so that we need appropriate methods to analyze this data, to extract health-related aspects from it.
10. 01:18And finally, of course, we need the appropriate interfaces, which then give the user a feedback, how his health status is and possibly motivate him to change something about it.
11. 01:31I want to address a topic today that is very important.
12. 01:38It's about signal quality, and I brought an example,
13. 01:44which was also a few years ago, but which shows very clearly how signal quality strongly influences our work and how important it is,
14. 01:58To ultimately get the health aspects out of the data, from the day-to-day data.
15. 02:05The example I brought once was a project, where it was about measuring health relevant data on the airplane.
16. 02:14Now in this day and age, we don't fly as much as we used to, but the health aspect is maybe even more important nowadays than it used to be.
17. 02:23In that sense, this issue is just as relevant as it was a few years ago, where it was mainly a matter of offering passengers who have health risks to provide appropriate health monitoring.
18. 02:38And it was specifically about looking at the ECG or the heart rate during the flight to look at, for example, to identify certain health problems.
19. 02:48But also to be able to measure, for example, whether some passengers are suffering from extreme fear of flying. and then to be able to help those passengers.
20. 02:58The big challenge is, as you can imagine, if you're going to build a system like this, that it has to be very comfortable.
21. 03:07Yes, no one wants to have to be fitted with quite a lot of cables and electrodes, so that he can get on an airplane,
22. 03:15but in the best case he sits down in the airplane, as he is used to it and doesn't even realize that you can then collect health data from him.
23. 03:26In general, if we look at the EKG, I think maybe everybody has seen that,
24. 03:32the so-called PQRSTU complex, so just the naming of the different waves in the ECG.
25. 03:41Now it's like this, when we talk about a medical ECG, so if we take electrodes that we place on the skin, to then derive the ECG, we get a perfect signal.
26. 03:53So that's a signal that you get when you go to the doctor, when you go to the hospital and there the ECG is recorded.
27. 04:00But now we want to record the ECG in our everyday environment, in our everyday activities, for example, on the plane, but also, for example, when we do sports.
28. 04:13And in such a system, in the ideal case, it would be, that the ECG electrodes are built into our T-shirt, for example, so that we wouldn't have to wear any additional equipment to collect that data.
29. 04:30But that means that the signal quality is then subject to very large fluctuations, because when we move, the sensors slip, because they are not glued to our skin and many other things.
30. 04:44That means we get data, some of which is good, that we can use to identify health issues... and other data that are completely noisy, that we can't use at all.
31. 04:56And that's exactly the problem and challenge that we had in this project, when it comes to measuring health data on airplanes.
32. 05:05We'll see how that was implemented here.
33. 05:09So there was an ECG system built into the back of the airplane seat, that was a so-called non-contact ECG system.
34. 05:17Which works fine as long as you sit still in the chair.
35. 05:22And here also exactly again the analogy to the sport driving, also there the measuring systems, if you have that in your shirt, work quite well, as long as you don't move too much.
36. 05:35If you move around a lot, then they slip, and then the measurement doesn't work anymore.
37. 05:39And what we end up with in all of these issues, Is a comfort signal quality dilemma,
38. 05:48which is always that if we have very comfortable sensors, it usually leads to low signal quality.
39. 05:59If we use sensors that are more accurate, it usually means, that they are not as comfortable,
40. 06:08that you just can't fit them into a chair or have them in your T-shirt as easily anymore, but that you have to stick something to your skin or strap it on.
41. 06:17And that there are but technical difficulties or simply, that it's not the modern day better,
42. 06:26So that we can't improve comfortable sensors in the sense, that they provide better data.
43. 06:32And the question, of course, is, yes, how do we deal with that?
44. 06:37How can we solve this dilemma, so that in the end we can measure health-related data from everyday life, without having to severely interfere with people's normal lives?
45. 06:52And the idea that you apply here is usually as follows: You build an additional model, an additional system that detects the signal quality.
46. 07:06So the idea is, if we can detect from the data that we are taking in, automatically detect whether we have good or bad data quality, then we can automatically focus only on the good data
47. 07:23and the data that ends up being just noise or is too bad, to be able to analyze them further, we can then leave them to the left.
48. 07:31And in order to make it work, so-called artifact sensors are used, so in the ECG example, you then use motion sensors, for example, to detect how much the person is moving at the moment.
49. 07:48and to ultimately take that motion information into the model, to allow the model to recognize when the signal quality is good or when is the signal quality bad.
50. 08:00In our airplane chair example, we solved it in such a way, that additional pressure sensors were added to the airplane chair,
51. 08:10pressure sensors, which measured whether we were now sitting quietly on the chair but whether we moved strongly back and forth on the chair.
52. 08:19And these artifacts or these auxiliary sensors were then very helpful to recognize whenever the signal quality was very good or when the signal quality was very poor.
53. 08:34Of course, you have to evaluate the whole thing accordingly.
54. 08:38One must then, if one builds such a thing, of course always show how well it works.
55. 08:43And that means, for each use case, you have to then the corresponding experiments and measure data,
56. 08:51in order to verify, in order to prove that the system is actually able to to detect the signal quality.
57. 08:58And that's what we did here in the airplane example.
58. 09:00And yes, what do you do on the plane?
59. 09:02We looked at that, there's also real statistics, which give information, what do people do in the airplane.
60. 09:08And then there are the five main activities: That's relaxing of course, that's watching TV or listening to the music in the board entertainment system.
61. 09:22That's reading, that's eating, and some are even working on the plane.
62. 09:26And exactly these different activities we ended up covering in this experiment.
63. 09:31So our subjects then sat on real airplane chairs, the technology was built in accordingly, And then performed these various airplane activities.
64. 09:41So that one can then show at the end, how well a health assistant like this actually works.
65. 09:50especially with regard to the signal quality, it is always very important at the beginning, that you compare it with measurements that are known to have a very high quality.
66. 10:06So one usually then takes medical equipment and record the data obtained from this medical equipment in parallel to the data that you have recorded with your own system.
67. 10:20And that's exactly what we did here.
68. 10:24So we have brought a medical ECG into use and measured the ECG at the same time in the airplane seat.
69. 10:33and then ultimately developed procedures on how to compare then, whether our airplane assistant measures as well or when it measures worse than the medical ECG.
70. 10:49And we then ultimately used this data that we obtained, to build exactly these models that can then detect, automatically detect, when is the signal quality from my health assistant good.
71. 11:05So when can I record, use the data density, To look at health related issues?
72. 11:12And when is better not to use these data, because they don't have the appropriate data quality to get appropriate benefit from it.
73. 11:21What we have used for this are procedures from machine learning, that is, methods that allow you to use the data that you have.., to show the system how it should react.
74. 11:36So building a system that actually learns from the data, in that case, learns what does low quality data look like,
75. 11:43and what does data with good data quality look like, in order to then later make this decision automatically itself.
76. 11:49This general approach that we have now seen in the airplane example, is an approach that can be found in many, many application areas.
77. 11:59I had already briefly alluded to it, when it comes to sports, we have the same problem, in many other areas as well.
78. 12:06We don't want to restrict people in their daily life, we want them to live the way they are used to.
79. 12:13And we just have the challenge then of trying to figure out when the signal quality is worse and when it still works well.
80. 12:25Finally, I have an example from a completely different direction, which, however, also illustrates normally how important it is to be able to work with sensors,
81. 12:39that are comfortable, that may not have the medical signal quality that we're used to seeing in medical applications, but that are comfortable to use in everyday life.
82. 12:53Now here's an example of an EEG system.
83. 12:57And maybe some of them have already seen how a medical EEG is measured.
84. 13:04When you look at pictures of the subjects, sometimes you have difficulties to recognize the subjects at all,
85. 13:09because they have many, many electrodes and many, many cables on their head, usually 32, sometimes even 64 electrodes distributed over the head.
86. 13:20Each electrode has a cable that then goes to the data recording system.
87. 13:25It usually takes half an hour or more to get the subjects first with this measurement system - to attach the measurement system to them.
88. 13:33Here is once an example, which shows, how one such a Headset, which also has EEG electrodes, can be used to help with speed reading.
89. 13:46So speed reading is a variation, where you can read a text very quickly.
90. 13:52You have to learn that, so instead of reading the text as we know it, with the whole document in front of us, here you get word by word presented one by one.
91. 14:02The challenge with this speed reading is that, once you set a speed,
92. 14:09that is, how fast the words are coming per second, it stays at that speed.
93. 14:14Now each of us knows that our concentration level just decreases from time to time, and it should actually be possible to measure that.
94. 14:25That is, when our concentration level goes down, this word speed should also be reduced.
95. 14:32And exactly that was done here now with a commercial EEG system, which is nothing more than a headset,
96. 14:41which you put on in one second and then you can measure the concentration level from the EEG information.
97. 14:48So this is far from being a medical EEG system, it has a much worse signal quality.
98. 14:54But, if you can develop the appropriate methods, to be able to assess that signal quality,
99. 15:02you can do great things with these devices, without having to tell people now that they have to attach quite a lot of electrodes in the case, for example, on their head.
100. 15:13Finally, an example from a very recent project:
101. 15:17It's about motion analysis, specifically it's about, measuring our step, our gait.
102. 15:25Interestingly, our gait says a lot about our health, not only about physical health, partly also about mental health, Keyword, for example, Parkinson's disease.
103. 15:37This is also a disease, that changes our gait behavior, our gait pattern.
104. 15:44And here the question is, to what extent can that be measured with sensors, That are in our shoes, for example,
105. 15:51that can measure our gait parameters in everyday life and can warn us whenever there is a change in our gait behavior.
106. 16:00and perhaps be able to recognize early on when a change is taking place, that we should take another look at.
107. 16:09Again, as homework, my question:
108. 16:15Now think about what technologies you would like to see, that is, new technologies that perhaps do not yet exist today,
109. 16:25which you would like to integrate, for example, into your clothing and everyday objects like now shown here in the shoes or in the headset of an EEG system,
110. 16:33what could be measured with these everyday objects still measure in health-relevant data,
111. 16:39what would you like to be able to measure, to make even more aspects of our daily life measurable,
112. 16:49In addition to what we already have now, in addition to our smart watches and other devices that are already on the market now.
113. 16:56Thank you for your attention to the second part,
114. 17:00and I look forward to to see you again in the third part.

To enable the transcript, please select a language in the video player settings menu.