A new machine learning model can predict the risk of progression from a clinically isolated syndrome, or initial event of multiple sclerosis (MS)-like symptoms, to clinically distinct disease, a study has found.
“In our study, we not only numerically estimated the risk of developing CIS patients, but also developed a machine learning model. [clinically definite MS] However, it also provides sufficient interpretability to understand the basis for that prediction,” the researchers wrote.
the study, “Predicting transition from clinically isolated syndromes to multiple sclerosis: an explainable machine learning approach” was published. Multiple sclerosis and related diseases.
In most patients, MS is characterized by flare-ups or flare-ups, in which symptoms suddenly worsen, followed by periods of remission, in which symptoms diminish or disappear completely. To establish a clinical diagnosis of relapsing MS, a person must experience multiple such attacks or have signs of brain damage occurring at multiple points in time. (Therefore, it is called “multiple sclerosis.”)
People who experience only one MS-like attack and have no other signs of MS are said to have clinically isolated syndrome (CIS). If a person with CIS subsequently has seizures or has evidence of new brain damage, the person may be diagnosed with definite MS. However, not all patients with CIS develop MS.
Predicting CIS to MS conversion presents challenges
Recent studies have shown that treating MS patients with CIS can reduce the risk of developing MS, but these drugs are expensive and carry the risk of side effects.
In theory, patients with CIS who are at highest risk of developing clinically definite MS should be best treated with more intensive treatments, while patients at lower risk should receive safer, less effective treatments. You may have to undergo treatment or wait for treatment. However, accurately predicting her MS risk in patients with CIS has proven difficult.
A team of Iranian scientists has created a machine learning model that predicts MS risk in CIS patients based on a combination of clinical and demographic data.
“To our knowledge, this study [machine learning] A model that uses a combination of MRI characteristics, clinical data, and demographic information to estimate the probability of CIS converting to CIS. [clinically definite] MS,” the scientists wrote.
In machine learning models, computers are trained to identify patterns in complex data using pre-established mathematical rules or algorithms. By identifying these patterns, computers learn how to look for similar patterns and understand future data.
To create the model, the researchers used a dataset of 273 Mexican patients with CIS who were followed for 10 years, during which time 46% developed clinically definite MS. Scientists trained the model using data from most patients and then tested its accuracy using data from the remaining patients.
This revealed promising accuracy. In testing, the model was able to accurately identify 75% of her CIS patients who developed MS and 81% of her patients who did not develop MS.
Training helps the model identify variables
Machine learning models are useful tools, but a common drawback is that it is difficult to determine which elements are most important in the model after training. To address this, the researchers conducted a series of experiments in which they retrained the model using different combinations of variables to find out what type of data the computer needed to make accurate predictions. I did.
“This approach increases the interpretability of the model and provides a clear understanding of how each variable influences the predictions,” the researchers wrote. “This is especially important in medical applications where understanding the reasoning behind the predictions is as important as the predictions themselves.”
This identified nine important factors, including demographic factors such as age, gender, and education, as well as specific features on MRI images and specific symptoms.
“These findings highlight the multifactorial nature of the disease. [MS] and the importance of considering a wide range of variables in predictive models,” the researchers wrote.
The research team said further studies using larger datasets are needed to validate the machine learning model.
