It was seeing a girl with diabetes that inspired medical doctor Eduardo Jorgensen to set up health tech company MedicSen, which applies AI solutions to diabetes monitoring. She believed she could not lead a normal life with her condition and Eduardo saw a way that AI could help patients like her find a better quality of life.
MedicSen is using AI and deep learning techniques to build a tool for diabetic patients, which tells them how their actions are affecting their body chemistry. The app recommends which actions the patient should take to deal with the changes.
Without the support tool, Jorgensen says diabetes patients cannot know what will happen if they drink a soda, or if they do more exercise than usual. “That is the uncertainty that a diabetes patient lives with,” he says. “[The MedicSen app] will tell them what happens to them specifically, to one patient, because what happens to you is different to what happens to someone else.”
Deep learning in smart diabetes monitoring
Jorgensen explains that the benefit of deep learning would be in training original mathematical models, rather than in developing new mathematical models based on deep learning.
“What we’re finding out is that all the machine learning algorithms, all the statistics we do in the end, they miss some things and you have to put in some really deep technology for learning in those algorithms that are able to correct the algorithm itself. And you don’t have to do it with your own analyst inside the company. So it’s a very good system to get more curiosity in predictive models, in diagnostics and in drug research.”
One of the big obstacles is access to data and a lack of quality data, which either needs to be researched or, says Jorgensen, you have to ask the patients.
Secondly, he says, the wearable sensors are not entirely accurate, so there is a margin of error on the measurements they produce.
“So if you build the algorithm that’s taking on those data points, they have some error from the patient in reality.”
The solution is to train those algorithms to understand those potential errors.
“Errors that are coming not from the patient, not from your system, they’re coming from what you’re using to measure. I think deep learning will help solve those problems. Training algorithms to further understand a specific situation that not only depends on patients but also on technology.”
“The way we train these algorithms is deep learning. We’ve created a biological model of the relations between different parameters that are important in diabetes. So we correlate food, exercise, stress, insulin, there are a lot of parameters that we get together. And once we have a general model, the general model gets trained with the patient data by itself.
“Once we have those individual models for the population, we make sure we understand the differences thanks to deep learning. And we grow the basic model from the specific individuals that use it.”
New technology techniques that use deep learning, such as AlphaGo and OpenAI, could be used in MedicSen’s code, says Jorgensen. “We’ll be able to use our own machine learning algorithm as a software agent that gets trained over different predictive models just to get to the optimal point. So I think deep learning is going to help us in getting accuracy.”
The needle-free smart patch and a non-invasive artificial pancreas
In the next few months, MedicSen will be turning all its tech and AI toward the general lifestyle management for chronic disease, not just diabetes. The plan is to start impacting different diseases like epilepsy, high blood pressure and some blood clotting diseases.
MedicSen is working on building a device that patients could “wear and forget about”, so the device makes decisions and takes action and the device would be personalised for each patient. It will be a needle-free syringe in the shape of a patch, rather like a nicotine patch, but for insulin or other drugs.
The main innovation in the patch is that it creates a micro-pore in the upper layer of the skin through which the insulin, say, is delivered.
MedicSen have validated the product in pre-clinical tests and are now making plans for clinical trials. Jorgensen says to expect the smart patch to come out at the end of 2019 or early 2020. Its first release will be a manual version, where the patient has to specify what needs to be delivered. In the next two years, the company are planning heavy clinical trials to validate the patch connected to the algorithm, which will be in effect a non-invasive artificial pancreas.