- Many organizations are trying to implement healthcare artificial intelligence (AI) projects using traditional infrastructure models, which isn’t working well, observed Pure Storage Global Healthcare Technology Strategy VP Josh Gluck.
By using a traditional infrastructure approach, organizations are running into scalability problems with their AI projects, Gluck told HITInfrastructure.com.
“We're finding that a lot of folks want to use AI, and they're trying to get the most out of their existing traditional infrastructures. Unfortunately, they're running into situations where it takes an inordinate amount of time to hone their algorithms. The results that come back have to be reinterpreted, or they have to rewrite the algorithm,” Gluck related.
One of the biggest AI pain points for healthcare organizations with traditional infrastructure is that they have data in isolated pockets throughout their IT environment. “Healthcare traditionally has built infrastructures for applications. It has not provided an infrastructure where the data can be accessed freely by research programs, analytics programs, or AI machine-learning programs,” he said.
A recent Gartner report backs up Gluck’s analysis. It found that the increasing complexities of AI and other advanced technologies could cause healthcare IT infrastructure to fall behind, resulting in a chaotic IT environment.
“As the advance of technology concepts continues to outpace the ability of enterprises to keep up, organizations now face the possibility that so much change will increasingly seem chaotic … The key is that CIOs will need to find their way to identifying practical actions that can be seen within the chaos,” said Gartner Fellow and Vice President Daryl Plummer.
Preparing Infrastructure for AI Projects
To successfully implement an AI project, healthcare organizations need to understand what data is available, where that data is located, and what data will be most useful for AI, Gluck said. The data should be moved to a shared accelerated storage platform accessible to those working on AI programs.
Gluck said that the most popular AI use case in healthcare right now is radiology. “What we're finding is that the majority of folks have a vast array of images in their PACS [picture archiving and communication system]. They have a wealth of information in terms of known data. They have the images and results,” he said.
PAC systems are typically used by hospital departments, such as radiology, cardiology, dental, and pathology, to manage digital images and share data.
“Now organizations are saying, ‘Look, we have this repository of information. We could use it to improve the outcome of a new patient when they come in. We could personalize treatment for them, especially in the cancer space. Let's use that data and let's put it to work,” he added.
The American College of Radiology Data Science Institute (ACR DSI) recently identified five use cases for AI tools in medical imaging: identifying cardiovascular abnormalities, detecting fractures and other musculoskeletal injuries, aiding in diagnosis of neurological diseases, detecting thoracic complications and conditions, and screening for common cancers.
“The ACR DSI use cases present a pathway to help AI developers solve health care problems in a comprehensive way that turns concepts for AI solutions into safe and effective tools to help radiologists provide better care for our patients,” said DSI Chief Medical Officer Bibb Allen Jr.
Gluck predicted that machine learning and AI will have a significance influence on healthcare over the next five to ten years.
“AI will help in selecting the therapy that makes the most sense for the patient. I think that you're going to see AI used more broadly to help reduce physician burnout because of the amount of data that physicians have to look at in order to treat patients,” he said.
“Machine learning and AI will be leveraged to support the physician in either diagnosing or treating patients based on the patient’s own data and data that the institution or organization has accumulated over time,” he added.
“Organizations need to understand that their data has a lot of value. If they change their strategy of building bespoke infrastructures on an application basis to one of building an architecture around their data, they'd be able to harness the data and get the value out of it, driving the organization forward,” he concluded.