WEBVTT
Kind: captions
Language: en

00:00:10.080 --> 00:00:16.240
My name is Alexandra and I'm a PhD student in the Helmholtz&nbsp;
Information and Data Science School for Health.&nbsp;&nbsp;

00:00:16.240 --> 00:00:20.080
So, I'm working both at the Karlsruhe&nbsp;
Institute of Technology (KIT) and the&nbsp;&nbsp;

00:00:20.080 --> 00:00:29.840
German Cancer Research Center (DKFZ) bringing&nbsp;
both together data science and health science.

00:00:31.520 --> 00:00:35.280
Imagine, someone is diagnosed with&nbsp;
cancer. Then there will be a CT scan&nbsp;&nbsp;

00:00:35.280 --> 00:00:42.320
made and I brought you an example of a lung&nbsp;
CT. You see three different orientations and&nbsp;&nbsp;

00:00:43.200 --> 00:00:48.000
the black parts here is the lung which&nbsp;
is easier to identify than other parts.&nbsp;&nbsp;

00:00:49.440 --> 00:00:54.160
Now, an expert has to identify where the&nbsp;
cancer is and where organs at risk are located.&nbsp;&nbsp;

00:00:54.800 --> 00:01:00.640
This is a very time consuming task if done&nbsp;
manually and can delay the treatment onset. So,&nbsp;&nbsp;

00:01:00.640 --> 00:01:05.600
I'd like to have a computer program that&nbsp;
can assist experts with this task. This&nbsp;&nbsp;

00:01:05.600 --> 00:01:10.240
is usually done with an artificial neural&nbsp;
network and you can see my network here.&nbsp;&nbsp;

00:01:12.240 --> 00:01:19.360
This looks like a text document but can do a whole&nbsp;
lot and it does already segment the tumor volumes.&nbsp;&nbsp;

00:01:19.360 --> 00:01:23.200
Unfortunately, not with a precision&nbsp;
that is applicable in clinics. So,&nbsp;&nbsp;

00:01:23.200 --> 00:01:26.000
I will tune all the parameters&nbsp;
and there's a whole lot of them&nbsp;&nbsp;

00:01:26.720 --> 00:01:31.040
to figure out the best settings.&nbsp;
Actually, it can help doctors.

00:01:37.760 --> 00:01:42.240
I like a lot of different things about&nbsp;
my job. Mainly my own project of course.&nbsp;&nbsp;

00:01:42.240 --> 00:01:45.680
I like the programming parts, I&nbsp;
do like the literature research,&nbsp;&nbsp;

00:01:45.680 --> 00:01:49.920
I do curate my own data and also supervise a&nbsp;
Hiwi (student assistant) and a Master's student.&nbsp;&nbsp;

00:01:50.720 --> 00:01:55.200
There are also some side projects I like&nbsp;
and this is for example the website I was&nbsp;&nbsp;

00:01:55.200 --> 00:02:00.560
programming with some group members. Also,&nbsp;
I'm writing a manuscript for school teachers,&nbsp;&nbsp;

00:02:01.440 --> 00:02:08.720
also organizing retreats and cooperations. Then&nbsp;
there are also some soft skill courses I can&nbsp;&nbsp;

00:02:08.720 --> 00:02:21.840
participate in or I do some science communication&nbsp;
days like "Girls Day" or "Science Week" at KIT.

00:02:23.680 --> 00:02:28.000
One of my greatest challenges that my&nbsp;
data labels show great variability.&nbsp;&nbsp;

00:02:28.560 --> 00:02:31.520
You can see two examples that I've drawn here.

00:02:34.080 --> 00:02:37.600
They are very diverse. So, a neural network&nbsp;&nbsp;

00:02:37.600 --> 00:02:42.320
can recognize patterns but if&nbsp;
there's no pattern within the data&nbsp;&nbsp;

00:02:42.320 --> 00:02:47.200
my neural network can actually not learn and&nbsp;
I need to overcome that challenge. My second&nbsp;&nbsp;

00:02:47.200 --> 00:02:53.280
challenge is that training times over neural&nbsp;
networks take several days. So, I need to wait.