Medilync Case Studies
Medilync was founded with the goal of revolutionizing the way diabetes is managed and treated in a value-based care landscape. In the value-based care landscape, a deep understanding of the patient and the ability to make recommendations about lifestyle choices are crucial to long-term success. These case studies, partnered with Microsoft, are to support that.
Leveraging Machine Learning for Better Patient Treatment
The initial version of the solution used a virtualized cluster environment running Hadoop on Linux, with machine learning done using Apache Mahout – a set of machine learning tools that deal with classification, clustering, recommendations, and other algorithms. Setting up and maintaining the Apache Mahout cluster had been difficult and time consuming. Additionally, Medilync needed a management interface for the cluster with provisioning scripts to get better insight about what was running on the cluster.
The ultimate goal was to do real-time analysis on the patient behavioral data that would enable the care provider to give feedback to the patient to help them improve their quality of life.
Overview of the Solution
In January 2015, Medilync developers spent several days with engineers from Microsoft’s Technology Evangelism and Development (TED) team. TED and Medilync engineers created an innovative solution that utilized Azure Machine Learning, Azure Event Hubs and Nitrogen (an IOT reference architecture implementation).
Medilync exacted their machine learning models using Mahout. When they moved their models to Azure ML Studio, Azure ML Studio enabled verification of each step of the process, from the earliest stages of training the model. This dramatically reduced the time to recreate and verify the models. Azure ML Studio also allows for running more than one model at a time, allowing Medilync to quickly and efficiently determine which models are most accurate by contrasting and comparing results.
Using IoT for Enhanced Glucose Monitoring
Medilync creates an all-in-one telehealth device, called Insulync. Composed of a glucose meter and insulin pen reader, Insulync enables doctors with the ability to interact with patients in a more impactful way, leveraging near-real time analysis of data feeds on blood sugar levels, insulin dosage, pulse, and blood pressure. Simultaneously, data collected via the device will be combined with meal plan data from partners such as MyFitness Pal and exercise information. All of this data is analyzed using machine learning to help understand and identify the best care options for a patient.
With the sensitivity of the personal data that is captured with the Medilync smart glucose meter, there are a number of challenges.
1. How to register and identify the devices as known devices and belonging to a given patient.
2. How to securely and reliably send data to the cloud at scale.
3. How to send commands and configuration to the device itself in a secure way.
With privacy concerns even more important when dealing with the personal information coming from the Medilync devices, solving these challenges are essential to the success of Medilync.
Overview of the Solution
In January 2015, Medilync developers spent several days on site at Microsoft with developers from the Technology Evangelism and Development (TED) team working to find a solution to these challenges. The solution to these three challenges comes in the form an open source project called Nitrogen. Nitrogen is a set of IoT components that leverage Azure and its premium services such as Azure Event Hubs, while providing security, scale and reliability, similar to the Azure IoT suite.
The smart glucometer that Medilync has created runs an embedded Linux implementation called OpenWRT which is capable of running node.js. Nitrogen has a small node.js agent that runs on the device to handle registration, authentication and communications with the server side.