The process

What do you have to do?

The goal of this project is to train a machine that diagnoses heart diseases with artificial intelligence. In order to do this, we will guide you through a set of steps, where you will have to choose a configuration for your machine. But don't worry, everything will be explained so you can choose knowing what option best suits you. Finally, you will have to upload a set of "solved" MRIs from your patients. That means MRIs the outcome of which is already known to you. For instance, an MRI of a patient who had a heart attack 6 monhts after the MRI was taken.

The Choices

The algorithms you take to set up your machine are based on your personal preferences, but for the sake of information we will explain how each one works. It's always better to choose if you have a proper understanding of why you're making your choices.

The use of the first algorithm that you will have to choose is to reduce the amount of data that your machine will have to process, because even if the machine would become more accurate by getting more data, it would take way too long to process it all. So what this algorithm will do is take the most important variables and get rid of the rest.
You can choose the "tactic" it uses to do so:·
·Correlation, that finds the variables that usually change together and groups them into one unique variable.
·Principal Component Analysis, which chooses the data that affects the result the most.
·Linear Discriminant Analysis, which finds a new dimension in which to fit different kinds of data as one.

The second algorithm that you choose will be the one that teaches the machine how to interpret the data that we give it. This way, when the machine is given a new MRI that isn't solved, it will have learnt from the ones that were used to teach it, and it will know how to diagnose the patient based on what it has seen in other MRIs.
For this step, you can choose between 4 optons:
·Support Vector Machine, which creates a threshold between groups of patients (ill/healthy, etc).
·Random Forest, that randomly creates many decision trees which "vote" their outcome, picking the one with the most votes.
·Neural Network, that immitates the structure of a human brain with several layers of "neurons".
·Generalized Linear Model, that classifies patients based on a logistic regression.

The MRIs

You may not have enough MRIs to properly train a machine. This can happen sometimes. We recommend that you train your own machine only if you have over 80 MRIs from patients willing to take part in the project. If you happen to have fewer than that, we have made available an option for you to download machines that other doctors have trained, so you can choose one of theirs if you lack enough MRIs.