There are 5 important steps when getting started with AI in the medical field. Enjoy the second part of the medical AI series. Haven’t read the first step yet? You can read it here.
I have my problem statement… Let’s roll!
Once the problem statement is defined, you have to create a solution space. A solution space is a catchy design term for ‘exploring a lot of ways to solve the problem, instead of just one’. To define the solution space, I like to use a metaphor. It may be a strange comparison, but starting an AI project is a little bit like making pancakes. Let’s assume you’re not the type of person who buys cheap packaged pancakes in the local store (because let’s be honest, they’re not that good). Instead you want to make them from scratch yourself. The first things you do, are finding the best recipe, searching for the right ingredients, and making sure you understand the different steps and proportions so you don’t mess up during the baking process.
Finding the right AI project ingredients
The steps in creating an AI project are really similar to those of baking pancakes, which is the main reason I use this metaphor. The first thing you have to do, is find the ingredients to reach your goal. In this case, one of the main ingredients is of course the data set. When talking to clients about availability of data, we see three common, yet different situations:
- They don’t have a data set, but are able to gather it by using their current product or service.
- They have a data set, but it’s unstructured and different by design. It’s dispersed in the company because they gathered it in the past to fulfill others, non-AI, goals.
- They already have a data strategy in place and the data is structured.
Based on one of those three starting points you can clearly see an individual track.
When there isn’t any data available, you need to organize a data gathering campaign. You need ingredients before you can start cooking. There are two options in this case. First of all, you can search for an existing data set. There are already a lot of great data sets available online, for free. Often these sets emerged from a research project or a past campaign. So there’s a fairly good chance to find one for your project. Most of the time, you’ll have need to change them because they were structured with another purpose or project in mind. Nevertheless, it’s always a great starting point.
If you can’t find an existing data set, the only choice is to create one yourself. How this should be organized, really depends on the use case. A general rule here is that you will need a lot of data so don’t limit yourself too much.
The importance of qualitative data sets
Once you’ve found a data set, the next question is: ‘Does every data sample of data have the right quality?’. This is important because when creating an algorithm, the quality of the algorithm will be based on the quality of the data. This is why we recommend to start with a quality check. If you notice some data samples aren’t of good quality, it’s better to consider creating a pre-filter. This pre-filter will automatically select the highest quality samples from your data set for further processing. This does mean that your data set should be large enough to leave enough data after filtering out the bad ones.
When creating an algorithm, the quality of the algorithm depends
on the quality of the data.
When the dataset has the right quality and is sufficiently large we can start cooking, right? No. Before you start, you still have to prepare all of the ingredients needed for a successful project. Every algorithm needs a platform on which it will work. An algorithm designed to be integrated in a wearable has other requirements then one that will be processed in the cloud or on a server. Another important ingredient is determining how future data gathering will take place and how you intend to retrain the model and update it in the field.
Most of the time, an AI project only can be successful if all stakeholders are involved not once, not only at the start or somewhere else down the line, but continuously and with a focus on the interdisciplinary aspects between them.
There’s are still a lot of things to consider. So if you’re interested in this topic, please check out our perspectives for more information.
Any questions on this post? Submit them below and we’ll get back to you soon!
