A diagnostic platform powered by generative adversarial networks,
using active reinforcement learning
to speed up training
and minimize the knowledge transfer
effort of human doctors.
We are currently active
in cardio, respiratory
and ophthalmology.
PROBLEM
01
NCDs
Non-communicable diseases
(NCDs) are the leading cause of death and disability worldwide (60%).
Among NCDs, cardiovascular, ocular and respiratory diseases account for 60% of deaths and 53% of total healthcare expenditures.
02
AI
Fully supervised
diagnostic AI suffers
from several limitations,
such as the need to train
on huge amounts
of labeled data and
its difficulties in managing inputs that are noisy, incomplete
or simply different
from the original dataset.
03
BARRIER
In a situation of shrinking budgets where 97% of resources are spent on acute treatment and not prevention, the adoption of telehealth technologies is hindered by a cost barrier, both in EU and in LDCs. AI can address this problem by supporting Medical Doctors in early detection of disease.
SOLUTION
OUR STORY
I'm a paragraph. Click here to add your own text and edit me. It’s easy. Just click “Edit Text” or double click me to add your own content and make changes to the font. Feel free to drag and drop me anywhere you like on your page. I’m a great place for you to tell a story and let your users know a little more about you. This is a great space to write long text about your company and your services. You can use this space to go into a little more detail about your company.
OUR VISION
I'm a paragraph. Click here to add your own text and edit me. It’s easy. Just click “Edit Text” or double click me to add your own content and make changes to the font. Feel free to drag and drop me anywhere you like on your page. I’m a great place for you to tell a story and let your users know a little more about you. This is a great space to write long text about your company and your services. You can use this space to go into a little more detail about your company.
TECHNOLOGY
I'm a paragraph. Click here to add your own text and edit me. It’s easy. Just click “Edit Text” or double click me to add your own content and make changes to the font. Feel free to drag and drop me anywhere you like on your page. I’m a great place for you to tell a story and let your users know a little more about you. This is a great space to write long text about your company and your services. You can use this space to go into a little more detail about your company.
SOLUTION
A semi-supervised learning AI
able to learn
from small labeled datasets augmenting their dimension,
and predicting possible variations or noise that could be met in the future, so to make diagnosis robust
in real world settings.
Arrhythmias & Vascular
We developed a software able to estimate the probability for the presence of several heart conditions such as atrial fibrillation, atrial flutter, other arrhythmias, hypertension, and abnormal heart rate variability.
Detection is performed analyzing the PPG trace created by video-selfie of user staring at the smartphone camera.
Respiratory
The software we are clinically testing is trying to estimate the probability for the presence of asthma and Chronic obstructive pulmonary disease. Detection is performed analysing the sound made by a user blowing into the smartphone's mic.
Ophthalmology
The software we are clinically testing is trying to segment OCT retina scans to highlight diagnostic signs for Diabetic Macular Edema.
Further tests will involve other ocular diseases.
Active Learning
Our platform is capable of selecting, from a clinical dataset, only the most informative samples, in order present them to human doctors in an iterative fashion and get their feedbacks.
This way, the knowledge transfer effort of the doctors is minimized, and AI training can go faster, saving up to 90% of time and costs.
HOW IT WORKS
The Active Reinforcement Learning platform detects the most informative samples of a clinical dataset and submits them to Medical Doctors for labeling.
Then, the GANs platform propagates labels from small, manually labelled image datasets to all the unlabelled ones, so that the diagnostic AI can be trained to identify conditions without
the need of costly and time consuming human labelling (semi-supervised learning).
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 876145.