Understand new medical images using artificial intelligence (AI) and existing reports
It’s not like reading X-rays or other medical images quickly and accurately
It is essential for the health of patients and can save lives.

This assessment depends on the availability of skilled radiologists,
As a result, quick response is not always possible.
For that reason, a PhD graduate at the MIT Computer Science and Artificial Intelligence Institute (CSAIL)
Ray Liao said, “We are able to reproduce the work of the radiologist every day.
I want to train the machine. “

MIT-led groups are trying to improve the ability of machine learning algorithms to interpret,
We’re using a vast radiology report that is incidental to medical burns.
The team also tried to increase the effectiveness of the approach by using the concept of information theory

We’re using it.
The mutual information is a statistical measure of the two interdependences.
The way it works is as follows.
First, the neural network is designed to provide a number of X-ray images of the patient’s lungs
Doctors are trained to determine the extent of diseases such as pulmonary edema.

That information is encapsulated within a numeric collection.
The separate neural network performs the same task for the text to inform the information
Indicates in a different numeric collection.

Then the third neural network maximizes the mutual information between the two data sets.
Integrate information between images and text in a coordinated way.
Professor Polina Goland, MIT, a senior researcher at CASIL, said

“The high inter-information between images and texts means that images predict text well
It means that the text predicts the image well. “
Liao, Goland and their colleagues offer some benefits
We introduced another innovation.

It’s not working through the full image and radiation reports.
Divide the report into each sentence and the image part associated with that sentence.
If you do this, Goland will

“We estimate the severity of the disease more accurately than we see the full image and the full report.
And because the model inspects smaller pieces of data, it can learn more easily
We can learn more samples. “

Liao feels that the scientific aspect of the project is attractive,
His main motivation is “clinically meaningful and applicable to the real world
It is to develop technology. “

Therefore, the machine learning model of MIT is currently available at Beth Israel Deaconess Medical Center
In particular, it is necessary to manage the decision making of heart failure patients in an emergency room environment

A pilot program is in progress to see how it affects your doctor’s methods.
According to Goland, the model can have very wide applicability.
“It can be used for all kinds of images and related texts in and out of the me

dical field.
Moreover, this general approach can be applied beyond images and text
It’s fun to think about it.”

By fonuder

Leave a Reply