At Imminenth, we create simulations of patient prognosis.
As a result, we can transform private patient data into practical medical insight. We are currently building "synthetic medical datasets" to unlock the potential of personalized and precision medicine.
What is Synthetic Data?
An easier way to access medical data for advanced analytics, automation, and machine learning.
Our simulations are powered by the use of bleeding edge machine learning algorithms such as variational autoencoders (VAE) and generative adversarial networks (GAN). Simply put, our simulations learn from a real datasets and creates cohort datasets that have two key features:
They retain their statistical properties for the purposes of any advanced analytics including neural network training
They are void of any patient information and are impossible to re-identify due to the mathematical irreversibly of the process
Preservation of Statistical Accuracy
Our process learns and encodes characteristics of real world datasets. The simulation is able to sample and rebuild the data; creating simulated patient prognoses. Essentially, our AI is rewarded on how realistic it can generate patient data. With these mechanisms in place, we are able to effectively reproduce high dimensional datasets with near perfect statistical accuracy.
Void of any Real Patients
We embed the concept of differential privacy within our generation process. Our simulations self monitor whether or not the generated dataset is meeting privacy standards. Essentially, our AI is punished if it generates a real patient. Once the creation of a synthetic dataset is complete, it becomes mathematically impossible to recreate the real dataset.
Uses for Synthetic Data
Our synthetic data can and should be treated as real data.
Build predictive models for mortality using traditional or deep learning methods
Analyze time-to-event data such as survival, outcome after intervention, duration of treatment, and much more
Model what-if scenarios based off of real world metrics
Determine underserved patient populations based on real and forecasted outcomes
Understand local prevalence of features, demographics, or disease for improved market intelligence
Apply classification algorithms to statistically categorize patient features and outcome
Share not only insights of the real data, but concrete examples without compromising patient privacy
Preserve data integrity and combine multiple datasets to maximize data insight discovery
Toronto, ON, Canada