https://mitchell-chatterjee.github.io/tags/automated-ecg/Automated ECGHugo Blox Builder (https://hugoblox.com)en-usMon, 01 Sep 2025 00:00:00 +0000https://mitchell-chatterjee.github.io/media/icon_hu3049596157658489136.pnghttps://mitchell-chatterjee.github.io/tags/automated-ecg/https://mitchell-chatterjee.github.io/project/automated_ecg/Mon, 01 Sep 2025 00:00:00 +0000https://mitchell-chatterjee.github.io/project/automated_ecg/<p>This is the official code repository accompanying the paper titled Toward Robust Automated Cardiovascular Arrhythmia Detection using Self-supervised Learning and 1-Dimensional Vision Transformers.</p> <h2 id="abstract">Abstract</h2> <p>This project developed a robust automated cardiovascular arrhythmia detection system using self-supervised pretraining and 1D vision transformer architectures tailored for ECG signals. The system emphasizes reliability under distribution shift and low-labeled-data regimes.</p> <h2 id="dataset--artifacts">Dataset &amp; Artifacts</h2> <ul> <li>Public datasets: (see repository for dataset details and preprocessing scripts)</li> <li>Code: </li> <li>Paper: Toward Robust Automated Cardiovascular Arrhythmia Detection using Self-supervised Learning and 1-Dimensional Vision Transformers</li> </ul> <h2 id="methods">Methods</h2> <ul> <li>Self-supervised pretraining on large unlabeled ECG corpora</li> <li>1D Vision Transformer fine-tuning for classification and detection</li> <li>Augmentation strategies, calibration, and uncertainty estimation for safer clinical deployment</li> </ul> <h2 id="results">Results</h2> <ul> <li>Achieved state-of-the-art robustness across held-out datasets.</li> <li>Demonstrated improved calibration and lower false positive rates in low-label settings.</li> </ul> <h2 id="relevance">Relevance</h2> <p>This work underpins my research focus on medical AI and time-series modelling, and complements my engineering work on large-scale retrieval systems by bringing rigorous evaluation and robustness-first design.</p>