Artificial intelligence could become the standard for diagnostics in hospitals, leading to new ways of assessing and diagnosing patients, researchers said.
In a paper published today in the journal Scientific Reports, researchers at the Massachusetts Institute of Technology (MIT) looked at how medical imaging systems work and how to create AI to automate and process data.
The researchers used an automated system that uses a neural network to analyze images and find patterns.
They then created a new AI that could process the data and determine how best to perform the analysis.
The system was able to recognize different parts of the body and recognize a patient’s health and pain using its own system.
“It’s really cool to see how machine learning is used in medical imaging, particularly with the potential for big data,” said the paper’s lead author, PhD student K.R. Khanna, who worked with MIT Professor of Computer Science, Professor of Electrical Engineering and Computer Science and the Director of the MIT Robotics Lab.
“This could lead to really powerful new technologies that will really improve healthcare in our lifetime.”
Using a neural net to analyze and identify data has been used in previous medical imaging work to improve diagnostic and treatment algorithms, such as image-processing algorithms that recognize tissue patterns and detect signs of disease in the body.
But, in this case, the team was able the to create an AI that can recognize parts of a person’s body and identify them based on a set of features.
The new system could be used in future diagnostic tools, Khanna said.
“We hope it will open up more capabilities for imaging and medical imaging.”
The team was also able to develop an AI for analyzing and extracting data from images that can perform different tasks.
It can be used to find different parts or features of a human body, including facial expressions and movements, he said.
The AI also could be trained to recognize patterns of data in images, like a pattern of blood or blood vessels, in order to analyze more details in a particular area.
For example, it could recognize a specific vein in a person, he explained.
This type of machine learning could allow doctors to better interpret the data to find the best possible diagnosis.
“What this means for us as an imaging system is that we can take a human subject and train an AI to recognize features that are associated with a specific body part, and that’s the same as having an actual trained AI, where you can actually use it in medical diagnostic tasks,” Khanna explained.
“If the doctor wants to say, ‘OK, we know that a vein has a red line running down it, that’s a red vein, and we want to see what’s going on inside that vein,’ you can do that with a trained AI.”
This type or neural network can also be used for more complex tasks, such in a diagnostic tool that could be able to learn how to identify multiple types of blood vessels in a blood vessel, he added.
The research is part of a larger effort by MIT researchers to develop medical diagnostic tools that would be able use machine learning to improve medical imaging in a way that will be more accurate and useful.
They are currently developing a new machine learning tool that can identify blood vessels from a human eye, as well as from the eyes of animals, and even the skin of people.
“There’s really been a huge push for this technology in medical science, and there’s been a tremendous amount of interest in this technology, and it’s been kind of a wild ride,” Khannasaid.
“I think there’s a lot of potential here.”
Khanna’s team was joined by Professor of Engineering, Professor and Director of MIT’s Computational Neuroscience Lab, Professor in the Computer Science Department, and Professor of Biological Engineering and Bioengineering.
The project is called DeepMind: Neural Image Processing.
Khannavas work has been funded by the Defense Advanced Research Projects Agency (DARPA), the National Institutes of Health (NIH) and the National Science Foundation.
Khana is also the first author on a paper in the Proceedings of the National Academy of Sciences (PNAS).
A video describing the work is available on YouTube.
For more MIT News, visit MIT News Center.