Chagas Disease Detection Using Optimized XGBoost and Explainable AI

Chagas disease is a silent threat that affects millions of people, particularly in Latin America. It is caused by a microscopic parasite that enters the body and can lie dormant for years, often showing no symptoms at all. The danger lies in what happens later: the parasite can slowly damage the heart, leading to irregular rhythms, heart failure, and a specific type of chronic heart muscle disease. Because the early stages are so quiet, catching the disease before serious heart damage occurs is incredibly difficult. Doctors rely on electrocardiograms, or ECGs, which are simple, non-invasive tests that record the heart's electrical activity. These tests are cheap and widely available, making them a perfect tool for screening, but reading them requires a trained eye, and subtle signs of the disease can be easily missed, especially in areas where medical resources are scarce.

In recent years, scientists have turned to computers to help spot these hidden patterns. While advanced computer systems known as deep learning have shown promise in analyzing heart signals, they often act like black boxes. They can make a prediction, but they cannot explain why they reached that conclusion. For a doctor to trust a computer's diagnosis, they need to understand the reasoning behind it. A team of researchers at Nilai University has developed a new approach that bridges this gap. They created a system that not only predicts the presence of Chagas disease with high accuracy but also explains exactly which parts of the heart's electrical signal led to that decision. Their work combines a powerful, efficient computer algorithm with a method that translates the computer's logic into clear, human-readable insights.

The researchers started with a large collection of heart data from a public database called CODE-15%. This dataset contained records from over 13,000 patients, each with a detailed list of measurements taken from their ECGs, such as the timing of electrical waves and the variability of the heart rate. To make this data even more useful, the team merged it with another file containing specific medical indicators, such as whether a patient had a blocked electrical pathway in the right or left side of the heart, or if their heart rhythm was normal. They then created new, custom features by combining these existing numbers in ways that might reveal hidden connections, such as looking at the ratio between different time intervals in the heartbeat. This process enriched the dataset, turning raw numbers into a richer picture of heart health.

With the data prepared, the team tested three different types of computer learning models to see which one could best distinguish between patients with Chagas disease and those without. They tried a simple, straightforward model, a complex model that builds many decision trees, and a highly advanced model known as XGBoost. The advanced model proved to be the most effective. To make it even better, the researchers used a specialized tool to automatically fine-tune its settings, running hundreds of trials to find the perfect combination of parameters. This optimization process allowed the model to learn more efficiently from the data, improving its ability to spot the subtle signs of the disease.

The results showed that this optimized system could correctly identify the disease in about 74 percent of cases and had a strong ability to separate sick patients from healthy ones. While other studies using massive, complex deep learning networks have achieved similar scores, this new system is far simpler and much faster to run. More importantly, the researchers did not stop at just getting a number. They applied a technique called SHAP to the final model, which acts like a spotlight on the decision-making process. This tool revealed exactly which factors were most important for the diagnosis. It showed that the patient's age was the single biggest clue, followed closely by the presence of a specific electrical block in the right side of the heart. It also highlighted that a normal ECG reading was a strong sign that the disease was absent.

The system went a step further by providing explanations for individual patients. For a specific person, the model could show that their high risk was driven by a combination of their age and a significant difference between their actual age and the age their heart appeared to be based on its electrical activity. This level of detail is crucial because it allows doctors to see the evidence behind the prediction, rather than just accepting a yes-or-no answer. The researchers found that the most influential factors, such as age and specific electrical blocks, align perfectly with what medical science already knows about how Chagas disease damages the heart. By combining strong predictive power with clear explanations, the team has created a tool that is not only accurate but also trustworthy enough to be used in real-world clinics.

This work suggests that sophisticated computer tools do not need to be mysterious to be useful. By focusing on a model that can be understood and optimized for efficiency, the researchers have offered a practical solution for early detection in resource-limited settings. The system is already available as a web application where users can input clinical data and receive an instant prediction along with a detailed explanation of the factors involved. While the study acknowledges that testing on even larger and more diverse groups of people would be valuable in the future, the current findings demonstrate that a balanced approach, one that values both performance and transparency, can effectively support the fight against this neglected tropical disease.