Scientists from the UK have presented a new method of using AI, which speeds up MRI by several dozen times. He will soon be taught to work with all types of scanners.
Researchers at King’s College London’s School of Biomedical Engineering and Imaging have automated the labeling of brain MRI images needed to improve machine learning models. They then recognize the images by extracting important clues from the radiology reports and assigning them precisely to the appropriate MRI examinations. Now more than 100 thousand pictures can be taken in half an hour.
Deep learning typically requires tens of thousands of tagged images to achieve the best performance in pattern recognition. This is the weakest part in the development of deep learning systems for complex imaging datasets, in particular MRI, which is fundamental for detecting neurological diseases.
“This model has greatly facilitated the tasks of deep learning image recognition, and it will almost certainly accelerate the advent of automated brain MRI readers in the clinic. The potential for patient benefit is enormous,” the researchers noted.
The study authors note that although one barrier to rapid research has already been overcome, there are further challenges to be addressed. Now scientists want to make their method work in most hospitals that use different scanners.
Previously, scientists at the University of California, Los Angeles used artificial intelligence (AI) to identify three new subtypes of multiple sclerosis. The researchers say their findings will help identify those people who are more likely to have progression of the disease and help predict treatment more effectively.