Saturday , July 24 2021

From bionic weapons to predict patient growth in ER, AI transforms patient care

TORONTO – Chris Neilson is the second prosthesis after she lost her left hand over the elbow, a six-year-old nasty job accident. Semi-myoelectric, artificial hand and arm are stepped cosmetically and functionally from the first body under the control of a device featuring a hook and a claw.

But the 33-year-old mining worker is looking forward to helping Albert's researchers investigate a new generation of prostheses – an experimental bionic arm that can "learn" to adapt and anticipate amputate movements using artificial force intelligence.

This is just one example of how AI, or more specifically, machine learning, is starting to transform healthcare into what was once a science fiction thing in reality.

Co-Director Patrick Pilarski and his research team at the University of Edmonton, Bionic, for improved natural control (BLINC), are developing an artificial hand using AI / machine training to attempt to disassemble lines between a person in need of assistive technology and support technology itself. ”

"They may have lost their limb injury or illness, and the technology itself is trying to replace this part of the body," says Pilarski, Canadian research manager for mechanical engineering.

"So, when we try to create better bionic extremities, are we really looking for how we understand the signals a person gives to their technology?" How does this technology interpret these signals to really do what they want to take a cup of coffee or take a pen or hold a sweetheart hand?

"And then how does this device give the person information so they can better fulfill their daily tasks?"

Artificial weapons that resemble the android Sonny additions to the 2004 film I, Robot are combined with the use of machine training software that takes amputate repetitive patterns, and then begins to incorporate these movements as the wearer controls prosthetics.

Using Neilson's current artificial limb, he manages hand-open and adhesion functions by closing the bicep or tricep muscles. Sensors measure electrical changes in the skin by sending signals to activate movements in the hand.

But with the BLINC laboratory prototype, "I could switch from the threshold, between the wrist or hand rotation to an open and closed hand, to maybe even change the handle," says Neilson of Ice, Alta.

"And what this software did if you had a model (movement) … it took what function you chose as a priority, and then after a while it skipped the rest."

Currently, the AI-enabled artificial limb remains in the testing phase, but the ultimate goal is to create a sophisticated prosthesis that the ampute can use in everyday life, says Pilarski, a colleague at the Alberta Machine Intelligence Institute, or Amii.

Knowing the manual control of an artificial limb, he explains that the idea of ​​a reinforced bionic arm in the AI ​​is to prevent some of the burden caused by the user, making more complex hand movements more automatic.

"This means that if someone is working very hard to give all these right signals to his part of the prosthesis, in such a way that it carries all the right movements, if the prosthesis in the limb can predict where they reach or how they want to grab the object, then it can deprive some of this overhead person.

"A person may have a more natural, intuitive and, in some cases, more effective interaction with his device, because the device fills them with gaps."

Throughout the country, efficiency is also partly over a dozen of AI-driven projects being developed in the United States. The Michael Hospital in the heart of Toronto, which serves a large number of city dwellers, is also one of the most important regional injury centers in Ontario. .

The Health Care Analytics Center Research Center (CHART) was created to develop and implement innovative programs using AI / machine training to streamline certain hospital systems and improve care. worse – and when.

CHART Director Dr. Muhammad Mamdani says that about twice a year the Emergency Unit is taking off for patients arriving through its doors, leaving staff encrypted and creating "ridiculously high" waiting times of 8 to 16 hours. Mini surges also occur a few times a month, which is also "pretty bad".

Therefore, the emergency department heads turned to the center to see if a program could be designed to predict the patient's volume, so more nurses could be transferred to shift, or the doctor's schedule changed.

Mamdani says that CHART staff looked at three-year historical data to identify ED usage patterns, then added environmental data: Will it deliver snow tomorrow night? Did the NBA Raptors play? Was there a marathon in the core of the city?

All the data was entered to create an algorithm – a set of rules that tells a computer how to perform a task – using a combination of techniques that included learning a machine, he says. "And we found that we can predict with more than 90 percent accuracy our patient volumes … at six-hour intervals, three days in advance.

"For example, if, on Monday, we can tell you that on Wednesday, at At 18.00 We will have 82 patients who will appear in our emergency department. We will be able to tell you that about 10 of them will have mental health problems, 12 will be high-intensity cases (heart attack, trauma), and others will be low to moderate. ”

The beauty of the program is that it is all automated, and the relevant data is taken daily, and the estimated amount of ED in seconds sent to the department heads.

CHART also addressed the concerns of internal medicine specialists, who know that some of their patients are dying or will have to be transferred to the IU – but doctors cannot predict which individuals are at risk.

"What they told us was that, until they realized that these patients were getting worse, they had an average of three hours to respond," said Mamdani. "They also said there were care routes that would reduce the risk of cardiac arrest and death by 40 to 50 percent," but we need about six to 13 hours to implement them. ""

That's why his team developed a deep learning algorithm that drives all kinds of data from patient records, including laboratory test results, medication, clinical orders, and more.

"We've read it in text notes, so looking for the words the sisters wrote last night, which a cardiologist posted in his or her morning," he said. “And if the patient dies or goes to the IU, it will take 12 to 24 hours before.

"As soon as it reaches a certain threshold, it speaks to our paging system, and it warns the medical team to see this patient within the next two hours because we think something is going to happen."

Program technology may seem complicated, but the goal is simple, says Mamdani.

"We hope it will really save life."

Source link