Scientists use AI to identify three subtypes of multiple sclerosis in MRI brain scans

According to foreign media reports, scientists may have identified three new subtypes of multiple sclerosis (MS). A team trained artificial intelligence algorithms on a large dataset of MRI scans of the brains of patients with multiple sclerosis and discovered patterns about which regions of the brain are first affected by the disease. This may lead to a more concentrated treatment plan for different types, or it may be a new treatment plan.

Multiple sclerosis is a common central nervous system demyelinating disease, in which the immune system mistakenly attacks the myelin sheath, which is a membrane that wraps around the axons of nerve cells. The function of myelin sheath is like the insulation layer on a wire, so when it is damaged, signals from the nervous system may be difficult to pass through. This manifests as symptoms such as muscle weakness, spasms, numbness, balance or coordination problems.

Multiple sclerosis is divided into four types based on the activity and progression of the disease. Clinically isolated syndrome (CIS) is the initial symptom onset, which may be a one-time event or may develop into MS; Recurrent remission type MS (RRMS) is a continuous stage in which patients experience alternating periods of symptoms and remission; Primary progressive MS (PPMS) is characterized by gradually worsening symptoms without any relief in between; Secondary progressive MS (SPMS) occurs in the later stages of RRMS patients, with remission stopping and a sustained decline beginning.

The problem with this classification is that these four types are more about different stages of the disease than different variants. They are not very useful in identifying which treatment options may be very effective for each patient, as patients may experience several types of them throughout their lives.

Therefore, for new research, researchers investigate whether there are other subtypes that may affect the path that patients are likely to take. If these can be diagnosed earlier, more professional treatment can be carried out, giving every patient good opportunities.

The team set up an artificial intelligence tool called "Subtype and Stage Inference" (SuStaIn) to perform MRI brain scans on 6322 MS patients. In doing so, it discovered three new subtypes of MS, which scientists later named "cortical dominant type", "normal appearance white matter dominant type", and "lesion dominant type". These names are based on where and what type of exception first appeared.

"At present, multiple sclerosis is roughly divided into the progression group and the recurrence group, which are based on the patient's symptoms and do not directly depend on the underlying biology of the disease. Therefore, it cannot help doctors choose the appropriate treatment method for suitable patients," said Dr. Arman Eshaghi, the main author of the study. "Here, we used artificial intelligence and raised a question: Can AI find subtypes of MS that follow a certain pattern on brain images? Our AI discovered three data-driven subtypes of MS defined by pathological abnormalities seen on brain images."

Next, the researchers locked in artificial intelligence and set it to work in MRI scans of an additional 3068 patients to verify its ability to recognize three new subtypes. From the data, the team is able to assign different attributes to each subtype. For example, in the dataset, the lesion dominant subtype is the least common, but it has a high risk of rapidly progressing disability.

After mastering this information, medical professionals can ultimately better diagnose patients and prescribe treatment methods that they are more likely to respond to.

"We conducted further retrospective analysis of patient records to understand how newly identified subtypes of MS respond to various treatment methods," Eshaghi said. "Although further clinical research is needed, according to subtypes, patients' responses to different treatments and the accumulation of disability over time vary significantly. This is an important step in predicting individuals' responses to treatments."

The study was published in the journal Nature Communications.

Prev: The Next Generation of Information Technology: The Spintronics Revolution May Come Soon