AI reads brain MRI images in seconds, automatically detects emergency cases

The system can identify multiple neurological diseases and assess the level of emergency requiring medical intervention. According to new research, this model achieves an accuracy of up to 97.5%.

According to researchers at the University of Michigan, Prima, the name of the AI system, is trained based on a large amount of actual MRI data, combined with patient medical records and clinical information. The results show that Prima not only recognizes diseases but also has the ability to classify cases that need priority treatment. The study was published in the journal Nature Biomedical Engineering.

According to Dr. Todd Hollon, the main author of the study and neurosurgeon at Michigan Medicine, the increasing demand for MRI is putting great pressure on doctors and the healthcare system. In that context, AI tools can help reduce the burden by providing faster and more accurate diagnostic information.

The research team tested Prima on more than 30,000 MRI scans with more than 50 different diagnostic imaging groups, including many serious neurological disorders. According to the research team, Prima provides higher diagnostic efficiency than many advanced AI models today.

The system can automatically identify cases that require high priority, such as stroke or cerebral hemorrhage, conditions requiring immediate medical intervention. In this case, Prima can send a warning to the appropriate specialist doctor to shorten the treatment time.

Prima belongs to the visual-language modeling group, capable of processing images and text simultaneously in real time. Unlike previous AI systems that were usually only trained for narrow tasks, Prima was trained on a large and diverse dataset, including more than 200,000 scans and 5.6 million image sequences, along with medical history and scan indications.

According to the author group, this approach helps Prima reason closely to clinical practice, when combining image data with the specific pathological context of each patient.

Every year, millions of MRI scans are performed globally, mostly related to neurological diseases. However, the rate of increase in MRI scan demand is far exceeding the capacity of diagnostic imaging experts, leading to overload and slow results delivery.

In many medical facilities, especially in areas lacking resources, patients may have to wait many days to receive diagnosis results. New technologies like Prima can contribute to improving access to diagnostic imaging services, regardless of the scale or location of the medical facility.

Despite achieving positive results, the research team emphasized that Prima is still in the initial evaluation phase. Subsequent studies will focus on integrating more electronic medical record data to improve diagnostic accuracy.

Dr. Hollon compares Prima to "ChatGPT for medical diagnostic imaging", with the goal of supporting and not replacing the role of doctors in clinical practice.

Read the original here