An explainable hierarchical machine learning framework for schizophrenia biomarker detection based on electroencephalogram signals

The human brain is a vast, humming network of electrical signals, constantly firing in patterns that shift with our thoughts, emotions, and states of health. For decades, doctors have relied on interviews and observation to diagnose complex mental health conditions like schizophrenia, a disorder that alters how a person thinks, feels, and interacts with the world. While these clinical methods are essential, they are also subjective, often leading to delays or misclassifications. In recent years, scientists have turned to electroencephalography, or EEG, a non-invasive technique that places sensors on the scalp to record the brain's electrical activity. Unlike expensive brain scans, EEG is affordable and portable, offering a direct window into the brain's real-time dynamics. The challenge has always been that these electrical signals are incredibly complex and subtle, making it difficult to spot the specific patterns that distinguish a healthy brain from one affected by illness.

A team of researchers from several universities in Iran has developed a new way to tackle this problem, creating a system that not only identifies these patterns but also explains exactly how it found them. By analyzing recordings from 101 people—51 diagnosed with schizophrenia and 50 healthy volunteers—the researchers built a machine learning framework capable of distinguishing between the two groups with notable accuracy. The study focused on two specific states: when participants sat with their eyes closed and when they sat with their eyes open. The system did not just guess; it sifted through thousands of mathematical descriptions of the brain waves, selected the most telling ones, and used a powerful computer algorithm to make its decision. Crucially, the researchers did not stop at a simple "yes or no" answer. They used a method to peel back the layers of the computer's decision-making, revealing which specific parts of the brain and which types of electrical activity were the most important for the diagnosis.

The core of this work lies in how the researchers handled the sheer volume of data collected from the EEG sensors. Each recording produced a massive amount of information, far too much for a computer to process efficiently without getting confused. To solve this, the team used a step-by-step selection process to find the twenty most useful measurements out of the thousands available. These measurements were not just simple averages; they included complex descriptions of how chaotic or orderly the brain waves were, as well as how much energy was present in different speed ranges of the signals. Once this smaller, high-quality set of data was ready, the researchers tested ten different types of computer learning models to see which one could best separate the patients from the healthy controls. They used a rigorous testing method where the computer was trained on data from almost everyone and then tested on a single person it had never seen before, repeating this for every participant to ensure the results were genuine and not just a lucky guess.

The results showed that the computer models performed best when the participants had their eyes closed. In this quiet, resting state, the most successful models correctly identified the condition about 81 percent of the time. The models that worked best were those that mimic how a group of experts might vote on a decision, combining many small judgments to reach a final conclusion. When the participants opened their eyes, the models were still able to distinguish the groups, but the accuracy dropped slightly, suggesting that the brain's internal, resting activity holds clearer signs of the disorder than its activity while processing the outside world. The computer did not rely on a single type of signal to make these judgments; instead, it found that a mix of different measurements worked best. The most important signals came from the front and sides of the brain, specifically involving the speed of the waves and the complexity of their patterns.

Perhaps the most significant contribution of this study is the transparency it brings to the process. In many previous attempts to use computers for medical diagnosis, the system acts as a "black box," giving an answer without explaining why. Here, the researchers used a tool to map out exactly which features drove the computer's decisions. They found that the most influential signal was a measure of complexity from the front center of the brain. When this signal was highly complex, the computer leaned toward diagnosing schizophrenia; when it was simpler, it leaned toward a healthy diagnosis. Other key signals included the strength of specific brain wave rhythms in the temporal regions on the sides of the head. This finding suggests that the disorder is not defined by a single broken part of the brain, but by a combination of changes in how different regions process information and how chaotic their electrical activity becomes.

The researchers were careful to note that while their system performed well, it was tested on a specific group of people from a single medical center. This means the findings are promising but not yet a final solution for every hospital in the world. The study explicitly argues against the idea that simple, small-scale tests are enough, showing that previous studies with very few participants often produced overly optimistic results that did not hold up when tested more strictly. By using a larger group of people and a stricter testing method, this study provides a more realistic picture of what is possible. The work demonstrates that combining a careful selection of brain wave measurements with advanced computer learning can create a reliable tool for diagnosis. It also proves that these tools can be made understandable, showing doctors and patients exactly which biological signs are being used to reach a conclusion. This approach offers a path toward more objective, brain-based tools that could one day help doctors diagnose schizophrenia earlier and more accurately, moving beyond subjective interviews to a clearer understanding of the brain's electrical reality.