Machine learning (ML) is causing quite the buzz at the moment, and it’s having a huge impact on healthcare. Payers, providers, and pharmaceutical companies are all seeing applicability in their spaces and are taking advantage of ML today. This is a quick overview of key topics in ML, and how it is being used in healthcare.
A machine learning model is created by feeding data into a learning algorithm. The algorithm is where the magic happens. There are algorithms to detect a patient’s length of stay based on diagnosis, for example. Someone had to write that algorithm and then train it with true and reliable data. Over time, the model can be re-trained with newer data, increasing the model’s effectiveness.
Machine learning on Azure
Machine learning is a subset of Artificial Intelligence (AI). AI can be thought of as a using a computer system to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. I won’t go into more detail on the distinction, but here are some resources to help you get started.
- Azure Machine Learning services enable building, deploying, and managing machine learning and AI models using any Python tools and libraries.
- Azure Data Science Virtual Machines are customized VM images on Azure, loaded with data science tools used to build intelligent applications for advanced analytics.
- Azure Machine Learning Studio which comes with many algorithms out of the box.
- Azure AI Gallery, which showcases AI and ML algorithms and use cases for them.
Examples of machine learning in healthcare
Current examples of initiatives using AI include:
- Project InnerEye is a research-based, AI-powered software tool for planning radiotherapy. This is authored by Microsoft Research.
- Healthcare.ai has developed several healthcare related algorithms that provide a myriad of insights.
Microsoft is continuing to commit resources to making healthcare more effective through ML. There are several programs and publicly available resources including:
- Microsoft Medical, health, and genomics in Microsoft Research.
- Intelligent health through AI and the cloud.
- Microsoft and Adaptive Biotechnologies announce partnership using AI to decode immune system.
All these initiatives are driven by algorithms developed by researchers, data scientists, developers and others. The accuracy of prediction or recognition depends on two factors: the data and features used to train the model, and the algorithm used to learn from that data. That’s why people in the ML/AI space are so interested in the many algorithms that can be used today.
Supervised and unsupervised learning
There are two types of algorithms, supervised and unsupervised. Supervised learning algorithms make predictions based on a set of examples. For instance, historical stock prices can be used to hazard guesses at future prices. Each example used for training is labeled with the value of interest—in this case the stock price. A supervised learning algorithm looks for patterns in those value labels. Supervised data makes predictions more precise because the model is being fed correct answers to learn about expected results.
In unsupervised learning, data points have no labels associated with them. Instead, the goal of an unsupervised learning algorithm is to organize the data in some way or to describe its structure. This can mean grouping it into clusters or finding different ways of looking at complex data so that it appears simpler or more organized. This form of training is less specific, and the people analyzing the output may not even know the right answers themselves. That said, unsupervised learning can provide great benefits when an algorithm is tuned properly to fill in the blanks. Algorithm tuning is a process of trial and error, facilitated by the Azure learning platform tools. Using experts to evaluate the results is also critical.
But the benefits are huge when looking at trends in a population. Take this question: Where in the United States do most people with multiple sclerosis live? This can lead to questions like, “why is this?” with ML, one can potentially find insights not observed through other business intelligence approaches.
Personally identifiable information (PII)
The data that comes from healthcare products and services like electronic health records can contain personally identifiable information (PII). Special consideration needs to be made for how an organization will use and treat PII data in a machine learning solution.
Usage example: diagnostic radiology
Consider the job of a diagnostic radiologist. These physicians spend a lot of time analyzing image after image to identify anomalies in patients and much more. They are often critical in making a diagnosis, and their decisions are based on what they find—for example, identifying a tumor.
AI can be used to assist a diagnostic radiologist. For example, Project InnerEye describes itself this way:
Project InnerEye develops machine learning techniques for the automatic delineation of tumors as well as healthy anatomy in 3D radiological images. This enables; extraction of targeted radiomics measurements for quantitative radiology, fast radiotherapy planning, precise surgery planning and navigation. In practice, Project InnerEye turns multi-dimensional radiological images into measuring devices.
The software assists the radiologist by automatically tracing the outline of a tumor. Radiology produces a large number of scans of an area (e.g. top to bottom of a brain). The radiologist typically goes through each scan and traces the outline of the tumor. After this is done, a 3D composite of the tumor can be produced. This task takes hours. Using ML, Project InnerEye does this in minutes.
Usage example: using machine learning to predict outcomes
Other very practical information can be forecast using ML and AI. For example, predicting the patient’s likely duration of stay in a hospital is a form of predictive analysis.
Predicative Analytics for nurses helps them take better care of patients and of themselves.
Wrapping up
Diagnostic and predictive analysis are today’s main benefits, but work is underway to take advantage of ML/AI in other medical problem spaces. Pharmaceutical and insurance companies are keen on ML right now because it helps them manage with data and identify core KPIs.
These technologies will fundamentally change healthcare. But with heavy regulations around the healthcare space, we’ll see incremental adoption of ML/AI as new benefits are uncovered, and new algorithms developed.
Recommended next steps
- Complete the Azure Machine Learning services quickstart. This quickstart will quickly get you started with Azure Machine Learning Services.
- Deploy the Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI - The Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI offers a turn-key deployment of an Azure PaaS solution to demonstrate how to securely ingest, store, analyze, and interact with health data while being able to meet industry compliance requirements.
Additional Resources
- Download Machine learning algorithm cheat sheet to help choose an algorithm.
- Read Welcome to Machine Learning Server for an introduction to Microsoft Machine Learning Server formerly named “R Server.”
- Review the Solution templates for Machine Learning Server for Industry-specific templates including one for healthcare.
- For an example implementation, download the code: Predict length of stay with the Machine Learning Server.