Screening for Chagas disease from the electrocardiogram using a deep neural network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Infectious Diseases

Background:

• Chagas disease (ChD) affects over 6 million people globally, often leading to severe cardiac complications in its chronic phase.
• Early detection of ChD is crucial for timely treatment and prevention of severe outcomes, yet current detection rates remain low.
• Electrocardiograms (ECGs) offer a non-invasive method for cardiac assessment, potentially useful for ChD screening.

Purpose of the Study:

• To explore the efficacy of deep neural networks (DNNs) in detecting Chagas disease (ChD) using electrocardiogram (ECG) data.
• To develop and validate a convolutional neural network (CNN) model for identifying ChD from 12-lead ECGs.
• To assess the model's performance in distinguishing ChD, including chronic Chagas cardiomyopathy (CCC), from control groups.

Main Methods:

• A convolutional neural network (CNN) model was developed using over two million ECG entries from Brazilian patients (SaMi-Trop and CODE studies).
• Model performance was evaluated on two independent external datasets: REDS-II (631 ChD patients) and ELSA-Brasil (13,739 participants).
• Performance metrics included Area Under the Receiver Operating Characteristic Curve (AUC-ROC), sensitivity, and specificity.

Main Results:

• The model achieved an AUC-ROC of 0.80 on the validation set and 0.68 (REDS-II) and 0.59 (ELSA-Brasil) on external validation datasets.
• For the ELSA-Brasil dataset, sensitivity was 0.52 and specificity was 0.77 when identifying ChD.
• When focusing on patients with Chagas cardiomyopathy, the model's AUC-ROC improved to 0.82 (REDS-II) and 0.77 (ELSA-Brasil).

Conclusions:

• Deep neural networks can detect chronic Chagas cardiomyopathy (CCC) from ECGs, but performance is weaker for early-stage ChD.
• The study highlights the need for larger, higher-quality datasets, as limitations were observed with self-reported labels in the CODE dataset.
• These findings offer potential for improved ChD detection and treatment, especially in regions with high disease prevalence.