Many improvements in health care, including detailed pictures of the brain and vivid images of cancerous lesions on organs, are the result of MRI. But access to such advances is limited predominantly to high-income countries because of the expense of the instruments and their high-power requirements. A cheaper, low-power alternative MRI instrument operates at a low magnetic field and uses simple permanent magnets. But its weaker signal—a consequence of signal strength scaling with the magnetic field strength squared—leads to blurry images, which obscure the anatomical detail that doctors need to identify when diagnosing and treating various diseases.
A research team, led by Ed X. Wu of the University of Hong Kong, has now developed an MRI device with a field strength 1/60th that of conventional MRI that can be plugged into a standard electrical outlet. Although the raw output is still blurry, the researchers developed a deep-learning algorithm to sharpen the images, and the results are similar to those of a typical high-field instrument.
Much of the cost of a standard 3-tesla MRI comes from its superconducting magnet and the RF shielding, which blocks external electromagnetic interference. By using neodymium permanent magnets and forgoing RF shielding, Wu and his colleagues developed an inexpensive 0.05 T instrument. Assuming it’s manufactured at an industrial scale, the hardware would cost about $22 000, which is far cheaper than conventional MRI machines.
To get a clear image from an instrument with a weak signal and no shielding, the researchers gathered data on electromagnetic interference from sensing coils inside the device during a normal scan and when no magnetic resonance signals were collected. The data, in addition to high-quality images produced by high-field MRI instruments, were used to train an artificial intelligence (AI) prediction algorithm that estimated interference-free MRI data. After training, the AI model successfully corrected the low-field instrument’s images, and at that point it doesn’t require any reference high-field images. As shown in the comparison below, the corrected images are similar in quality to images from a high-field instrument.
Wu and colleagues aren’t the first to use AI to process blurry images from low-field MRI instruments. But their results are better than previous AI techniques. And even without RF shielding, their device can limit electromagnetic interference enough to clearly image multiple anatomical structures, including the brain, spine, and heart. Because of the low-field instrument’s affordable price tag and meager power demand, the use of MRI could expand greatly. MRI access in the US, for example, is roughly 40 scanners per million people, but the average across all African countries is only about 0.7 scanners per million people. (Y. Zhao et al., Science 384, eadm7168, 2024.)
