Digital cardiology faces a persistent challenge: electrocardiogram (ECG) records that reach artificial intelligence systems are often incomplete. Whether due to short display formats, lead loss, signal corruption, or simply having been digitized from images, these fragmented data limit the ability of models to make accurate diagnoses. In this context, the development of deep learning architectures such as ImputeECG represents a significant advance, enabling the complete reconstruction of 12-lead, 10-second ECGs, preserving observed samples and filling in missing regions with high fidelity.
ImputeECG is based on a mask-conditioned Transformer autoencoder, an approach that combines the efficiency of sequential models with the ability to attend to long contexts. Trained on the PTB-XL repository and validated on datasets such as CPSC2018 and a real clinical cohort of 43,633 records from the Kailuan study, the model reduces the mean absolute error in missing areas by 41 to 51%, and improves the reconstruction of critical parameters such as the R peak, RR interval, QRS duration, and P, T waves and the QRS complex. This not only restores the electrical signal but also recovers diagnostic utility: in multi-label classification, completed ECGs achieve an AUROC of 92.28% and an AUPRC of 33.88% in the most extreme loss scenarios.
Behind these capabilities lies a real need to process massive volumes of clinical data, much of which comes from analog archives or legacy systems. For such a solution to work in production, a robust infrastructure is required that combines artificial intelligence, AWS and Azure cloud services, and a custom software approach that adapts models to the particularities of each hospital center. At Q2BSTUDIO, we understand that innovation in digital health cannot be limited to the algorithm; it must integrate with cybersecurity systems that protect sensitive data, custom applications that connect with clinical workflows, and business intelligence services such as Power BI to visualize reconstructions and their impact on decision-making.
Furthermore, the ability to convert incomplete records into AI-ready signals opens the door to deploying AI agents that continuously monitor signal quality and authorize real-time completion. Q2BSTUDIO's experience in AI for businesses and in developing artificial intelligence solutions adapted to critical environments allows us to accompany healthcare institutions throughout the entire process, from ECG digitization to the implementation of inference pipelines that improve diagnostic accuracy without adding burden to medical staff.
Ultimately, ImputeECG demonstrates that the reconstruction of incomplete biomedical signals is not just an academic exercise, but a practical strategy to extend the useful life of ECG archives and enable more complete digital cardiac assessments. With the right support in terms of cloud infrastructure, cybersecurity, and application development tailored to each organization, this technology can become a clinical standard. At Q2BSTUDIO, we work to ensure that every piece of the ecosystem—from the model to the user interface—is aligned with the real demands of medical practice.
