Construction and Multi-Model Performance Evaluation of an Intelligent Screening Model for Meige Syndrome Based on Machine Learning

For decades, doctors have faced a frustrating puzzle when patients arrive with involuntary twitching of the eyelids or grimacing of the mouth. These symptoms, which define a condition known as Meige syndrome, often look exactly like common eye irritations such as dry eye or inflammation of the eyelids. Because the outward signs are so similar, patients frequently receive treatment for the wrong condition, leaving the true neurological disorder untreated for years. The core difficulty lies in distinguishing a specific type of muscle spasm that originates in the brain from simple eye problems, a task that currently relies heavily on a specialist's experience rather than clear, objective tests. Without a reliable way to tell these conditions apart quickly, patients miss the window for early intervention, such as targeted injections that can stop the spasms before they cause lasting damage.

A team of researchers at Zhengzhou University and the Third People's Hospital of Henan Province has tackled this diagnostic challenge by building a new kind of screening tool. Instead of looking for complex chemical markers in the blood, which require time-consuming lab work, they turned to the information a doctor gathers simply by asking a patient questions. By analyzing the medical records of nearly 4,000 people, the researchers trained computer systems to recognize subtle patterns in routine clinical history that humans might miss. They tested four different mathematical approaches to see which one could best sort patients into three distinct groups: those with Meige syndrome, those with other eye diseases, and those who are completely healthy. The goal was to create a fast, easy method that a general practitioner could use to decide if a patient needs a specialist's attention.

The study began by gathering data from 3,996 individuals, carefully divided into three equal groups to ensure a fair comparison. One group consisted of patients confirmed to have Meige syndrome, another included people with common eye conditions like dry eye or blepharitis, and the third group comprised healthy individuals with no neurological or eye issues. The researchers extracted twelve pieces of information from each person's file, ranging from their age and gender to their lifestyle habits like smoking and drinking, as well as their medical history regarding conditions like thyroid disease, sleep quality, and other issues affecting the face or head. They then split this data, using most of it to teach the computer models how to recognize the differences and saving a smaller portion to test how well the models performed on new, unseen cases.

When the researchers compared the four computer models, they found that one approach, known as an artificial neural network, outperformed the others. This type of model is designed to mimic the way the human brain processes complex information, allowing it to spot intricate connections between different symptoms. In the final test, this model correctly identified the group a patient belonged to about 77.7% of the time and showed a high ability to distinguish between the three conditions. While another model called a random forest came in a close second, it tended to be less consistent when predicting the most likely cases. The other two methods, which rely on more traditional statistical formulas, did not perform as well at this specific task. The results suggest that the neural network is the most reliable tool for this particular job.

A crucial part of the research involved simplifying the tool to make it practical for everyday use. The team discovered that the full list of twelve questions was not necessary; just five specific factors carried the most weight in making an accurate diagnosis. These five key indicators were the patient's age, the quality of their sleep, their history of drinking alcohol, whether they had a history of thyroid disease, and whether they had any other conditions affecting the face, eyes, nose, or brain. By focusing only on these five items, the researchers created a streamlined version of the model that performed almost as well as the complex version using all twelve variables. This simplified tool requires no blood tests and can be completed in under a minute, making it feasible for use in primary care clinics where time and resources are limited.

The findings also shed light on the specific characteristics that differentiate these groups. The data confirmed that Meige syndrome is most common in women between the ages of 55 and 60, a demographic that showed a much higher prevalence of the condition compared to the healthy group. Sleep quality emerged as a significant factor, with patients suffering from Meige syndrome reporting poor sleep far more often than those with eye diseases or healthy individuals. Interestingly, the study found that patients with Meige syndrome were less likely to have a history of drinking alcohol compared to healthy adults, a difference the researchers attribute partly to the older age and female predominance in the patient group, though they noted that alcohol can sometimes temporarily relieve symptoms in certain movement disorders. Another strong differentiator was the presence of other craniofacial conditions; while these were rare in healthy people, they were surprisingly common in the group with other eye diseases, helping the model tell the difference between a simple eye issue and the more complex neurological disorder.

The researchers were careful to note that their work has limitations. Because the data came from a single hospital and was collected by looking back at past records, the results need to be confirmed by testing the model on patients from different hospitals and in real-time settings. Additionally, some of the information, such as stress levels and sleep quality, relied on patients' own descriptions, which can be subjective. Despite these constraints, the study demonstrates that a simple, non-invasive screening tool is possible. By using a computer model trained on just five routine questions, doctors may soon have a practical way to identify Meige syndrome early, ensuring that patients receive the correct diagnosis and treatment much sooner than they do today.