KIST, CT images can be obtained without radiation exposure
AI technology development. Without opening the skull, using only MRI.
Transcranial focused ultrasound treatment expected to become possible


Transcranial focused ultrasound is a technology that can treat degenerative movement disorders, intractable pain, and mental illness by delivering ultrasound energy to specific areas of the brain without opening the skull. This treatment must be used in conjunction with imaging-based technology that identifies the location of brain lesions. In order for the focus of the ultrasound passing through the skull to be precisely focused on the brain lesion, information about the patient’s skull, which is difficult to identify with MRI alone, had to be obtained through CT.

CT, which inevitably exposes one to radiation, poses safety concerns for patients who need frequent examinations, children, and pregnant women. On the 2nd, Dr. Hyungmin Kim’s team at the Bionics Research Center of the Korea Institute of Science and Technology (KIST) announced that they had developed an AI technology that generates CT images based on MRI images and conducted a mock treatment experiment, confirming that transcranial focused ultrasound treatment is possible using only MRI. The results of the study were published in the latest issue of the international academic journal in the medical field, ‘IEEE Journal of Biomedical and Health Informatics (JCR(%): 1.8)’.

Although there have been efforts to obtain skull information from MRI images, special MRI coil equipment and imaging techniques that are not widely available in the medical field were required. Another alternative, AI-based CT image acquisition technology, has also attracted global attention, but has not been proven to be clinically effective. The research team proved that CT images obtained by AI can be used clinically through simulated treatment using ultrasound.

The KIST research team developed a 3D conditional adversarial generative network model that learns the process of nonlinearly transforming CT from T1-weighted MRI images, one of the most widely used images in the medical field.

We designed a loss function to minimize the HU (Hounsfield Unit) pixel variation error of CT images, and optimized the performance of the neural network by comparing the changes in synthetic CT quality according to MRI image signal normalization methods such as Z-score normalization and piecewise linear histogram matching normalization.

In order to provide stable and efficient ultrasound treatment, the patient's skull density ratio and skull thickness must be identified in advance. When these skull factors were identified using synthetic CT, both factors showed a correlation of over 0.90 with the actual CT and there was no statistical difference. In addition, when simulated ultrasound treatment was performed using synthetic CT, it was confirmed that the ultrasound focus distance error was less than 1 mm, the ultrasound sound pressure error was approximately 3.1%, and the focus volume similarity was approximately 83%. This demonstrates the possibility of performing transcranial focused ultrasound treatment using only MRI images.

Dr. Hyungmin Kim of KIST said, “Patients will be able to receive focused ultrasound treatment without worrying about radiation exposure, and the additional imaging and alignment process for medical staff will be simplified, which is expected to reduce time and financial costs.” He added, “We plan to develop it so that it can be applied to various treatment technologies through follow-up research to identify error rates according to ultrasound parameters and transducers and the possibility of applying AI CT to various parts of the human body.”