In my previous blog, I outlined some of the technologies that make up artificial intelligence (AI) and addressed the massive digital transformation that is taking place in healthcare today. In this follow-up post, you’ll see several use cases for AI technology in healthcare.
The first use case and a common starting point for AI is population health. Using predictive models built on classic machine learning or cognitive systems, healthcare providers can use characteristics from current patients to predict patients at risk for acquiring chronic conditions, not adhering to care plans or repetitive readmissions. Predicting these events before they occur allows for earlier intervention to reduce the cost of care.
A second use case for AI is machine learning based models in the clinical setting.
Examples of common predictive models include using electronic medical record data to predict hospital-acquired conditions like sepsis, operational models that predict patient admission rates into the emergency department, or financial models that can be used to identify new bundled service offerings.
A specific example of a predictive clinical model is a Rapid Response Team model that Intel developed with Sharp Healthcare in the United States through a joint proof of concept.
This model uses data from the electronic medical record to predict which patients are at highest risk for needing an intervention from a rapid response team, also known as a medical emergency team. Hospitals could use this model to proactively position rapid response team members and equipment to decrease the response time. Using historical data to test the model, we found about 80 percent accuracy in predicting patients who would require a rapid response event, with a slight bias towards false positives over false negatives.
With this particular model, Intel will be open sourcing both the code and an implementation guide later this year so other organizations can build a similar model with their electronic health record data.
A third use case for AI in healthcare is the application of deep learning to analyze medical images. This is an area where Intel has partnered with industry and providers in using deep learning on medical images for automated tumor detection.
The deep learning space is rapidly evolving. New frameworks and use cases are emerging regularly. This use case, in particular, is an area where healthcare can learn from applications developed by other industries.
Similar solutions can be applied to sports analytics and other important workloads.
A fourth use case for AI is with virtual services, like telehealth.
Increased adoption of telehealth has led to a proliferation of enterprise and consumer solutions, ranging from inpatient telehealth robots like those from InTouch Health to tablet-based telestroke solutions being piloted in ambulances.
These solutions generate rich video datasets, which could be used to develop AI solutions to improve clinical decision making. For example, in the telestroke use case, deep learning based models could be used to identify stroke signatures earlier in patients to improve diagnosis and potentially reduce the time to respond to treatment.
The fifth and final use case for AI is creating next generation virtual reality assistants.
In the future, AI could be used to auto generate specific elements of a virtual reality simulation in advance or in response to participant interactions in the virtual reality session. In the patient experience use case, the patient could interact with the virtual environment and see how pathology might change. For surgical training, AI could be used to analyze images to identify best practices from top surgeons that could be fed back into simulations to improve them over time.
In closing, digital transformation is creating amazing new opportunities for healthcare providers.
Through this process, organizations should look to embrace the use of data as a core capability to improve business process and patient experience.
As organizations get more advanced in their analytics abilities, the adoption of AI will naturally follow.