We need you to detect epileptic crisis on examples of real data (EEG signals), along with the crisis already detected by a human as an example to calibrate your algorithm (this will be considered the "groundthruth"). You can do that with the programming language you prefer, Matlab or C# preferred, but not a mandatory. For this task we give you data about where that crisis are, in fact, we give you the solution because what we need is the automatic process to detect more crisis on new EEG files in the future.
## Deliverables
**Task Name:** Detection of crisis in EEG signals
**Summary:**
We need you to detect epileptic crisis on examples of real data (EEG signals), along with the crisis already detected by a human as an example to calibrate your algorithm (this will be considered the "groundthruth"). You can do that with the programming language you prefer, Matlab or C# preferred, but not a mandatory. For this task we give you data about where that crisis are, in fact, we give you the solution because what we need is the automatic process to detect more crisis on new EEG files in the future.
Figure 1 - Polyman EEG Signal visualization (in the attached files)
This is the visualization of the EEG signal with Polyman, the software that you can download from [[login to view URL]][1] (freeware). You must study the signal and deduce the system to automatically detect the same crisis you can find on [login to view URL] file we attach with the description.
As you can see on the next image, we mark on yellow the same period of crisis on the signal representation and [login to view URL] registry. On this way you will be prepared to analyze the rest of the crisis and prepare the software which automatically takes this decision for us.
You can manage [login to view URL] as you please but the output we need from your software would be almost the same of [login to view URL], so that we can automatically compare how well your algorithm is performing. Will be something like:
# Start End
1 17h03m35s 17h03m40s
Figure 2 - [login to view URL] file (in the attached file)
**Deliverables:**
We will set the 1st deadline in **one week**. On this date you should give us the first deliverable that must consist on the software needed. In this moment we will test your solution against a new .edf file looking at the results and inform you about the results obtained. If the accuracy is less than 90% we'll let you recalibrate your algorithm and have a second chance.
If accuracy detecting crisis in the final algorithm is lower than 2 seconds, we will pay only 10% of the money, since the algorithm is useless for us (assuming no false negatives, and false positives lower than 20% of total crisis).
If the error is lower than 1 second you will be paid 100% and between 1 second and 2 seconds 50%.
** **
**Concepts to take into account:**
False negatives: When a crisis is occurred and you don't detect it.
False positives: When you detect a crisis and it never occurs.
Artifact: As you can see, not all the visually important changes in the signal are crisis, you have to take into account the grounthruth we provide.