Comprehensive coverage

How can the length of time subjects lie down in an MRI machine be shortened?

An artificial neural network can fill in missing information in imaging images of organs such as the brain

Preparation for MRI examination. Image: depositphotos.com
Preparation for MRI examination. Image: depositphotos.com

An MRI (Magnetic Resonance Imaging) is a scan that uses a powerful magnetic field and radio waves to produce accurate (and even 3D) images of the organs being examined and the blood vessels for diagnosis or medical research. The magnetic field and the radio waves affect the spin of the protons in the nuclei of the hydrogen atoms found in the water molecules in the body. The imaging is done by sending a series of impulses (pulses) of radio waves that create magnetic resonance and thus cause the protons to leave equilibrium and move to a higher energy state. After each pulse ends, there are decay and energy emission processes in which the protons return to their initial state and produce a movement that causes an electromagnetic signal, and the input coils of the device receive such a signal for each pulse. To get accurate and detailed images, many signals are required, so the series of pulses is long. This is why an MRI scan takes a relatively long time.

Acquisition of resonance images
Acquisition of resonance images

Prof. Tami Ricklin Raviv from the School of Electrical and Computer Engineering at Ben Gurion University deals with the processing and analysis of medical and biological images, especially MRI images of the brain. For this purpose, it uses tools from the field of artificial intelligence, such as artificial neural networks. These networks are a computational mathematical model developed inspired by the neuron networks that exist in the human brain. Similar to the brain, they "acquire intelligence" through learning from examples and thus can perform a variety of computational tasks. They consist of layers (input, output and intermediate) of many information units (neurons) that are linked to each other and transfer numerical data from one to another. In the learning process, the numbers that represent the strength of the connections between the information units are updated. These networks can be used in almost all computer applications, including deciphering biological and medical simulations and improving their quality. Acquisition of resonance images

"The main part of my research deals with the analysis of medical imaging images - for example, detecting tumors and pathologies (for example in brain cells) that cause aging, Alzheimer's, neurological and psychiatric diseases. That's why many of my research partners are clinicians and neuroscientists," explains Prof. Ricklin Raviv.

The latest research by Prof. Ricklin Raviv and her team, which won a grant from the National Science Foundation, aims to find a way to shorten the process of acquiring images of organs (such as the brain) in MRI. According to her, "Patients, including children, adults and patients, are inside the MRI machine for a long time, which causes discomfort and claustrophobia, and in order to produce clear and unblurred images, they must lie still and sometimes even hold their breath for a long time. That is why we wanted to find a technology that would reduce the number of pulses of the radio waves, that would speed up the process of acquiring the images but at the same time preserve their quality. For this purpose, we decided to use the artificial neural networks, which can, through learning, supplement information that the MRI machine did not acquire."

Reconstruction of resonance images
Reconstruction of resonance images

The researchers took brain images produced by MRI, and reduced some of their resonance frequencies. In this way they created images that contain only part of the original information, as if the subjects were inside the device for only about a quarter or a fifth of the usual time and each brain scan they underwent lasted about a minute instead of five minutes. At the same time, they developed a dedicated artificial neural network, which was trained using deep learning methods, to complete the removed frequencies.

The researchers took brain images produced by MRI, and reduced some of their resonance frequencies. In this way they created images that contain only part of the original information, as if the subjects were inside the device for only about a quarter or a fifth of the usual time and each brain scan they underwent lasted about a minute instead of five minutes. At the same time, they developed a dedicated artificial neural network, which was trained using deep learning methods, to complete the removed frequencies to obtain high-quality images. To make sure that the neural network completed the correct information, the researchers examined the reconstructed images compared to the original ones and calculated the difference between them according to accepted metrics (formulas based on the signal-to-noise ratio, which, among other things, are used to calculate the compression quality of digital images). which will reproduce resonance images

"Radiologists are often afraid of the possibility that the neural network will come up with information that did not exist (such as a tumor) or alternatively ignore or mask essential clinical information. Therefore, to test the clinical reliability of the reconstructed images, we examined them from partial scans of patients with multiple sclerosis, a disease that also manifests itself in (relatively small) lesions in the white matter of the brain. In the verification process, we separated the lesion areas from the healthy brain tissue in both the reconstructed and original images, compared the results and showed that the differences are very small. From this it can be understood that even if the restored image does not look exactly like the original, the doctor can obtain from it the clinical information he needs - such as the location of the lesions, their volume, their borders and their condition - and that is enough, this is the correct measure to use. The idea is to restore the full frequency picture without losing clinical information or 'inventing' it - that's the challenge," concludes Prof. Ricklin Raviv.

Life itself:

Prof. Tami Ricklin Raviv, married + four children, lives in Ramat Gan. Aims to increase the proportion of women in the subjects of exact sciences and engineering, especially electrical engineering ("I would be happy to see more female students with us").

More of the topic in Hayadan: