Elias Stenhede

• Stenhede, Elias; Orstad, Eivind Bjørkan; Omland, Torbjørn; Schirmer, Henrik & Ranjbar, Arian (2026). Associations between echocardiographic traits and artificial intelligence-enabled electrocardiography predictions of heart failure. The European Heart Journal – Digital Health (EHJ-DH). 7(8). doi: 10.1093/ehjdh/ztag141. Full text in Research Archive
• Stenhede, Elias; Ravn, Jesper; Schirmer, Henrik & Ranjbar, Arian (2026). Heart failure detection in electrocardiograms using Artificial Intelligence and pragmatic labelling. npj Digital Medicine. doi: 10.1038/s41746-026-02774-4. Full text in Research Archive Show summary

  The diagnosis of heart failure (HF) is resource-intensive, leading to severe underdiagnosis. This study proposes the use of a deep learning model to detect HF solely from electrocardiograms. HF has limited validity in diagnosis codes, lowering the viability of direct application of supervised learning. However, validating the codes with measured levels of circulating N-terminal proB-type natriuretic peptide (NT-proBNP) during training mitigates the impact of label noise. We developed and prospectively validated a neural network for HF detection, using a development cohort of 25,300 patients and an independent test cohort of 43,727 patients. The model achieved an AUC of 0.86 (95% CI 0.847–0.864) for hospital-diagnosed HF, which increased to 0.91 (95% CI 0.900–0.913) with age-adjusted NT-proBNP thresholds and 0.96 (95% CI 0.951–0.963) with stricter limits of 125 and 1000 ng/L. The model was externally validated using the MIMIC-IV cohort of 161,352 patients, yielding AUC values of 0.87 (95% CI 0.861–0.871), 0.90 (95% CI 0.895–0.908), and 0.96 (95% CI 0.950–0.964) with the same labelling strategies. Model predictions were further validated using echocardiographic assessments of diastolic function grade and the H2FPEF score, demonstrating that the model accurately captures both diastolic and systolic function. The model has been released as open source.

• Reyna, Matthew A.; Deepanshi; Weigle, Audrey T.; Koscova, Zuzana; Campbell, Kiersten & Shivashankara, Kshama Kodthalu [Show all 17 contributors for this article] (2026). ECG-image-database: large-scale paired ECG images and time-series with real-world artifacts; a foundation for computerized ECG digitization and analysis. Physiological Measurement. ISSN 0967-3334. 47(7), p. 1–22. doi: 10.1088/1361-6579/ae85b2. Full text in Research Archive
• Johansson, Elias Stenhede; Bjørnstad, Agnar Martin & Ranjbar, Arian (2026). Digitizing paper ECGs at scale: an open-source algorithm for clinical research. npj Digital Medicine. 9. doi: 10.1038/s41746-025-02327-1. Full text in Research Archive Show summary

  Billions of clinical ECGs exist only as paper scans, making them unusable for modern automated diagnostics. We introduce a fully automated, modular framework that converts scanned or photographed ECGs into digital signals, suitable for both clinical and research applications. The framework is validated on 37,191 ECG images with 1596 collected at Akershus University Hospital, where the algorithm obtains a mean signal-to-noise ratio of 19.65 dB on scanned papers with common artifacts. It is further evaluated on the Emory Paper Digitization ECG Dataset, comprising 35,595 images, including images with perspective distortion, wrinkles, and stains. The model improves on the state-of-the-art in all subcategories. The full software is released as open-source, promoting reproducibility and further development. We hope the software will contribute to unlocking retrospective ECG archives and democratize access to AI-driven diagnostics.

• Johansson, Elias Stenhede & Ranjbar, Arian (2025). Biomarker-Based Pretraining for Chagas Disease Screening in Electrocardiograms. Computing in cardiology. ISSN 2325-8861. 52. doi: 10.22489/cinc.2025.298. Full text in Research Archive Show summary

  Chagas disease screening via ECGs is limited by scarce and noisy labels in existing datasets. We propose a biomarker-based pretraining approach, where an ECG feature extractor is first trained to predict percentile-binned blood biomarkers from the MIMIC-IVECG dataset. The pretrained model is then fine-tuned on Brazilian datasets for Chagas detection. Our 5-model ensemble, developed by the Ahus AIM team, achieved a challenge score of 0.269 on the hidden test set, ranking 5th in Detection of Chagas Disease from the ECG: The George B. Moody PhysioNet Challenge 2025. Source code and the model are shared on GitHub: github.com/AhusAIM/physionet-challenge-2025.

• Johansson, Elias Stenhede; Singstad, Bjørn-Jostein & Ranjbar, Arian (2024). ECG-DUaL: Pose-Invariant Digitization of Printed Electrocardiograms via U-Net and Local Edge Detection. Computing in cardiology. ISSN 2325-8861. 51. doi: 10.22489/CinC.2024.262. Full text in Research Archive Show summary

  ECG recordings are often stored as paper scans, due to clinical accessibility and tradition. This paper presents a digitization pipeline for ECG printouts based on line detection to correct for rotation, and a U-net for segmentation. The digitization step is differentiable, so the Unet can be fine-tuned for any evaluation metric of interest. The tool achieves a maximum translation-adjusted SNR of 3.047 dB on the George B. Moody PhysioNet Challenge 2024 hidden dataset, placing Ahus AI Lab fourth among the official entries

View all works in NVA

• Johansson, Elias Stenhede; Ravn, Jesper; Schirmer, Henrik & Ranjbar, Arian (2025). Detection of hospital-admitted heart failure regardless of ejection fraction using artificial intelligence-enhanced electrocardiograms. European Heart Journal. ISSN 0195-668X. 46. doi: https:/academic.oup.com/eurheartj/article/46/Supplement_1/ehaf784.1167/8310259. Full text in Research Archive Show summary

  Heart failure (HF) presents with diverse and often non-specific symptoms, frequently overlapping with those of other conditions, making it challenging to diagnose accurately. Developing a more affordable and widely available diagnostic method that provides immediate feedback could improve the care and management of patients with HF. Previous work on AI-enabled ECG for detecting HF has largely been based on echocardiographic labelling for model training, and often focused on particular phenotypes of HF like HF with reduced or preserved ejection fraction (EF). Our aim was to develop an AI model identifying HF across the EF spectrum and to show that explicit echocardiographic labelling is in fact not necessary, and that a cheaper data labelling strategy results in AI models with desirable diagnostic properties. ECGs from a network of four tertiary hospitals in the period 2016-2024 were used, with data from 2016-2022 for development (40,000 patients, 161,000 ECGs) and 2023-2024 (43,000 new patients) for prospective testing. Using our proposed labelling strategy based on ICD-10 codes and NT-proBNP levels in the development set, 20,000 ECGs were labelled as HF and 58,000 as non-HF, while 83,000 ECGs were excluded due to uncertain status i.e. NT-proBNP between 125 and 1000 ng/L. An ensemble of cross-validated AI models was evaluated on the first ECG of each of the 43,000 patients (3,500 with HF diagnosis), including 5,000 who had an echocardiogram within 14 days. The model achieved high diagnostic accuracy for hospital-diagnosed heart failure (HF) with an AUC of 0.84 (95% CI 0.831-0.848), regardless of EF and NT-proBNP levels. For the population with measured NT-proBNP levels, the model outperformed NT-proBNP in diagnostic accuracy (p = 7.5e-7). In the test set with echocardiography, the model detected HF with reduced EF with an AUC of 0.91 (95% CI 0.894-0.931) and HF with preserved EF with an AUC of 0.68-0.89, depending on definition. Comparison with the H2FPEF risk score showed that model-predicted risk increased with the score and further stratified risk within each category. To highlight the impact of underdiagnosis in evaluating the model, we conducted a small-scale retrospective clinical evaluation of 60 patients with EF >50%, no HF diagnosis, and normal NT-proBNP levels (<125 ng/L). Out of the 60 patients, 30 had a low model-predicted risk while 30 had a high risk; 24 of the 30 highest-risk patients met HFpEF criteria, while 27 of the 30 lowest-risk patients did not. Our study shows that using a dataset of patients with clinical HF or no HF validated via NT-proBNP in the development phase, a neural network can be trained to detect of HF independent of EF, including HFpEF with low NT-proBNP. The model could serve as an accessible diagnostic marker for HF in primary care, as a screening tool to identify patients who may benefit from further evaluation by e.g. echocardiography.

• Stenhede, Elias; Ranjbar, Arian & Schirmer, Henrik (2026). Trustworthy AI-ECG for heart failure detection. Universitetet i Oslo. ISSN 9788234808760. Full text in Research Archive Show summary

  Heart failure has a worse prognosis than most cancers, yet despite the existence of effective treatment, the condition often goes unnoticed with current diagnostic tools. This thesis shows that artificial intelligence applied to electrocardiograms (AI–ECG) can detect heart failure both with reduced and preserved ejection fraction. This work focuses on the development and validation of AI–ECG models for heart failure detection and on tools that convert paper ECGs into digital signals. Together, they enable the use of AI in healthcare settings with varying degrees of digital infrastructure. The AI–ECG model showed strong performance across both internal and external cohorts, and the AI–ECG risk of heart failure was closely related to echocardiographic measures of cardiac structure and function. The strongest correlation was found for global longitudinal strain, suggesting that the model might detect early myocardial dysfunction. All models and digitisation software have been released openly to support reproducibility and clinical use. Together, this work establishes AI–ECG as a scalable tool that can complement current diagnostic pathways and enable earlier detection of heart failure.

View all works in NVA