CT images generated from MRI alone
Image credit: Dreamstime
Researchers from the Korea Institute of Science and Technology (KIST) have developed AI technology for producing CT images based on magnetic resonance imaging (MRI). This approach could allow clinicians to obtain detailed information about their patients without radiation exposure.
Transcranial-focused ultrasound is a non-invasive treatment used for degenerative movement disorders, pain, and mental disorders. In order to target a specific area of the brain, the treatment must be applied with an image-based technology for locating brain lesions.
Doctors typically use computed tomography (CT) to obtain information about a patient’s skull that is difficult to identify with MRI alone. CT scans expose the patient to radiation doses, which, while generally considered safe, are much higher than the doses associated with conventional X-rays. The number of CT scans could be lowered if more detailed information could be extracted from lower-dose imaging such as MRI.
Now, researchers from the Bionics Research Centre at KIST have developed an AI tool to generate CT images based on MRI images. A simulation experiment showed that a transcranial ultrasound procedure could be performed using MRI alone, using this approach.
Previous efforts have been made to obtain cranial information from MRI images, but these require specialist coils for the MRI or imaging protocols that are not widely available in medicine. Although there has been research interest in obtaining CT images from MRI, their clinical efficacy has not been proved; this research demonstrates that AI-based CT images have clinical utility.
The KIST team developed a 3D conditional adversarial generative network – a machine learning approach often used for generating images – that learns the non-linear CT transformation process from weighted MRI images. They optimised its performance by comparing change in quality of the generated CT images according to the normalisation methods of MRI image signals.
Safe, effective transcranial-focused ultrasound treatment requires knowledge of the patient’s skull density ratio and skull thickness. These factors, when obtained via the synthetic CT, showed over 90 per cent correlation with the real CT and no statistically significant difference. When simulated treatment was performed, the ultrasound focal distance had an error of less than 1mm, the intercranial peak acoustic pressure had an error of 3.1 per cent, and the focal volume similarity was 83 per cent.
“Patients can receive focused ultrasound treatment without being worried about radiation exposure,” said Dr Hyungmin Kim of KIST. “And as the additional imaging and alignment processes can be omitted, this will reduce the staff’s workload, leading to a reduction in time and economic costs.”
“Through follow-up studies on identifying the error associated with the ultrasound parameters and transducers and understanding the possibility of [AI] CT application in various parts of the body, we plan to continue developing the technology for its applicability in various treatment technologies.”
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.