Transformer AI detects early heart disease in ECG study
A 1D Transformer model identified early heart disease from ECG and clinical data with accuracy reported at up to 94.2%.
By Tom Brennan · Health & Medicine Correspondent
2 min read
A Transformer-based machine-learning model detected early signs of heart disease in electrocardiogram data with accuracy reported at up to 94.2%, according to research published in the International Journal of Medical Engineering and Informatics. The work points to a possible aid for clinicians reading ECGs, though the authors said it still needs further testing before use in live care.
The study, by Amal Miloud Aouidate, examined a one-dimensional Transformer network for heart disease detection using ECG signals and clinical data, according to the journal record. Inderscience, which provided the report, described the system as using an architecture first developed for language-processing tasks and applying it to medical signal analysis.
According to Inderscience, the model was tested with data from several well-known medical datasets. The report said the system analyzed ECG traces alongside other clinical information and showed strong performance in identifying early-stage heart disease.
Heart disease remains a major health burden, with almost 18 million premature deaths each year attributed to it, according to the report. Inderscience said early detection is a central challenge because treatment is more likely to help when cardiovascular disease is found before it advances.
ECGs record electrical activity in the heart and are widely used in diagnosis, according to the report. Inderscience said interpreting those traces takes clinical expertise, can take time and carries a risk of error, which has made automated support tools a focus of research.
The study’s reported accuracy does not mean the model is ready to replace medical judgment. Inderscience said the approach would be used with expert clinical assessment, potentially giving health care teams another source of evidence when deciding whether a patient needs further diagnosis or treatment.
The researchers said more development is needed, along with validation on independent clinical datasets, before the model can be assessed in a live clinical environment. The paper was published in 2026 under the title “Heart disease detection using 1D transformer network: case of ECG signals and clinical data,” according to the International Journal of Medical Engineering and Informatics.
This story draws on original reporting from Medical Xpress.