AI Helps Detect Rare Heart Condition from Scans

A new approach could help doctors diagnose a rare and dangerous heart condition using artificial intelligence. Brugada Syndrome (BrS) is an inherited disease that can cause sudden cardiac death, often in men around age 41. Currently, diagnosis relies heavily on spotting a specific pattern in electrocardiogram (ECG) readings. However, many hospitals still use printed ECG records rather than digital files, making automated analysis difficult. This research demonstrates how AI can bridge that gap by converting scanned ECGs into digital data and then identifying BrS patterns.

Researchers developed a system that can detect Brugada Syndrome from scanned ECG images. The method involves two main parts: first, digitizing the printed ECG signals, and second, using a machine learning model to classify whether the ECG shows signs of BrS. The system was tested on ECG leads V1 and V2, which are key indicators for this condition. In experiments, the AI model achieved an area under the curve (AUC) of up to 0.79 on validation data, indicating good ability to distinguish between BrS-positive and healthy ECGs.

The process begins by taking scanned ECG images and converting them into digital time-voltage signals. The pipeline handles different types of ECG printouts—newer color-grid versions, older monochrome ones, and binary images without grids. It automatically rotates crooked scans, removes background grids, and extracts the heart signal data. This digitization step transforms the visual waveforms into numerical data that computers can analyze. The researchers used Python with libraries like OpenCV and NumPy to build this pipeline.

Analysis of the results shows that the digitization successfully reconstructed ECG signals in most cases, preserving the waveform integrity. However, when signals overlapped strongly or image quality was poor, some inaccuracies occurred. For the classification part, the team used a Long Short-Term Memory (LSTM) neural network, a type of AI suited for time-series data like heartbeats. The model was trained on 30 BrS-positive examples from digitized scans and 80 healthy examples from a public database. It learned to recognize patterns associated with BrS, particularly in leads V1 and V2.

This technology matters because Brugada Syndrome is rare and can be missed, leading to fatal outcomes. By automating the analysis of existing ECG printouts, it could provide a second opinion to physicians, especially in regions where digital ECGs are not standard. The method also makes old ECG records usable for research, potentially uncovering undiagnosed cases. For patients, it means a quicker, more accessible way to identify a serious heart condition.

Limitations include the small dataset of BrS examples, which affected model performance. The digitization process sometimes faltered with low-quality scans or overlapping signals, and the AI model did not achieve perfect accuracy. The researchers note that further validation with medical experts and larger datasets is needed before clinical use. Future work could improve signal separation and explore combining this approach with other AI techniques for better reliability.