ECGLight: Compute-Light Framework For Paper ECG Digitization and Myocardial Infarction Screening

Imagine you're in a remote clinic where patients come in with paper ECG printouts, but you lack the internet or computing power to analyze them with modern AI tools. This is a common issue, leading to missed diagnoses of serious conditions like heart attacks because the technology just isn't available where it's needed most. Current solutions often require heavy computational resources or high-speed internet, which aren't always feasible in these settings. This is what's called a resource bottleneck.

To tackle this, the authors developed a system that can take a simple photo of a paper ECG and convert it into a digital format that can be analyzed on a standard smartphone or computer without needing a powerful server. The system not only digitizes the ECG but also screens for critical heart conditions like myocardial infarction, all while being quick and efficient. It uses a method called SHAP to help explain its decisions, making it easier for doctors to trust the results.

What sets this work apart from previous efforts is its focus on creating a complete, lightweight solution that works in low-resource environments. By training and validating the system on a large dataset of ECGs, it achieves impressive accuracy rates, meaning that even in places where technology is limited, healthcare providers can still access reliable diagnostic support. This could significantly improve patient outcomes in remote areas where timely medical intervention is crucial.

The paper introduces a novel lightweight on-device pipeline for ECG digitization and diagnosis, addressing a significant gap in current methods.

The claims are supported by extensive evaluation on multiple datasets, demonstrating solid performance metrics.