How will machine learning and artificial intelligence change medical informatization?

Medical information technology has always been a difficult point in the information process, before the development of medical information technology and the future trends, in this article, the reader will see how a group of engineers in the United States to use artificial intelligence and machine learning to promote medical information.

Steven Horng, an emergency doctor and engineer at Beth Israel Deaconess Medical Center in Boston, is a member of the department that aims to introduce machine learning and artificial intelligence (AI) to the emergency room.

"A lot of people are talking about how to get a computer to diagnose a patient or replace a doctor, but I think it's the wrong direction, and it's actually an enhancement to the doctor, a tool, not a substitute." ”

Horng's 10-person team is introducing a technology that is almost ignored by colleagues. The machine learning algorithm is hidden under the workflow, helping the medical staff to operate in a more effective manner. To touch it means developing new data science, running highly agile experiments, and relying on the flexible infrastructure of Beth Israel Chief information Officer Ohn Halamka, a component expert who points out that this is the core of the enterprise's introduction of machine learning and artificial intelligence.

Data Science project

HORNG's research team is using machine learning algorithms to address the data quality problems that plague emergency rooms across the country: a structured, standardized way to learn about the causes or "major claims" of patient visits. Medical personnel need to be precise, which makes data collection a battle. Use words to record the main appeal, for example, to show pain in the left chest.

"It's very valuable to be able to confirm chest pain for subsequent clinical care and things like sequencing and clinical pathways," Horng said, "but it's hard to achieve in most systems." "The difficulty has always been how to collect such structured data when people speak from their own words." ”

Horng's team and their computer science partners at New York University understand that if humans and machines want to work together to build unstructured data in practical applications, the system needs to do two things. According to their paper predicting figuratively complaints at Triage times in the Emergency Department (prediction of major claims during triage at the emergency department), "Users must feel that the software does save their time, And the results are credible. ”

Introduce machine learning. "We've got a machine learning approach that transforms these unstructured data into structured data domains, both at the back end and the front end," Horng said. Now, when a patient is being diagnosed by a nurse, the data collected is passed through a predictive analysis engine, which determines that the top five patients are most likely to be the main appeal.

"We have to build a very broad set of basic data--from the ontology of how to perform the main appeal to the natural language process," he said. This includes deciphering the "messy" parts of unstructured data: spelling mistakes, double meanings and tricky "negative discovery" issues, or reliable identification of medical conditions. Today, Horng says, the plan has dramatically increased the data-collection rate of the main appeal-from 25% to 95%.

A dedicated research and development team

Horng, who spoke highly of Beth Israel's Halamka, gave the team "the right to explore freely so that we could play a meaningful role in information technology support," he said. Horng and his team are members of the IS Information Systems Technical Support Department, adhering to the policy, procedures and governance structure of the information systems technology support is, and directly controlling hardware and software. "This is very different from most organisational physicians as advisers and hindsight," he said.

Kenneth Brant, a Gartner analyst, argues that CIOs should have a kind of learning when it comes to building smart-machine strategies and using machines to learn to complete the technology that replaces human tasks. Setting up a special group to carry out research and development is a good starting point for the cultural preparation of the business, Brant said. In addition, building a team like this will help CIOs think about the skills needed to drive the smart-machine strategy – those that might not be part of it.

"How to build smart machines or manage or deploy them is very different from [running] ERP systems," Brant said. "It requires an artificial intelligence skill that is deeper than the IT knowledge base." ”

Because intelligent machine technology is still immature, enterprises need to be able to test and test, and to accept the rapid failure of employees.

Horng and his team hold a fast-losing mindset. "We believe in the doctor programmer approach, where users are both doctors and developers and testers," Horng said. "This means we have a very fast development cycle. ”

To build the future, we must think flexibly

Infrastructure construction is also critical. HORNG relies on the flexibility of the Halamka architecture, which is hailed by experts as a way to ensure that the infrastructure can keep up with the changing state. Beth Israel Server virtualization has multiple failsafe coverage, Horng said. Horng and his team are running machine learning projects on a dedicated server-independent of the rest of the clinical environment. "This means that whatever happens to machine learning-if it crashes, or if a problem arises, or if the server fails-it doesn't affect the clinical aspect," he said.

Citrix Startup Accelerator chief technology expert Michael Harries, calling it a "turning point" in the architecture, is a concept he believes is more important than ever. "We are at a point in time where the total amount of change in the industry has been quite significant over the last decade, but the reality is that we haven't seen anything yet," he said at the new technology conference EmTech, sponsored by MIT Marvell Review MIT Technical review.

"I can centrally control all applications, whether on a mobile phone or on a desktop computer, or on any terminal that will appear, or where I can access it from anywhere," he said. "The delivery of content and it delivery should be separated." ”

(Responsible editor: Mengyishan)