Abstract 16137: Cardiac Implantable Devices Identification Based on Chest-x-rays Using Machine Learning Algorithms

• Published in: Circulation (New York, N.Y.) N.Y.), 2020-11, Vol.142 (Suppl_3 Suppl 3), p.A16137-A16137
• Main Authors: Melendez, Dorys A C, Lowey, Seth, Dastmalchi, Lily N, Nguyen, Tran, Adam, Gina, Mercader, Marco
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.142.suppl_3.16137

IntroductionHigh-quality care of patients with Cardiac Implantable Electronic Devices, involves early identification of device manufacturer, which leads to timely interrogation using specific programmers. Pacemaker ID is a phone application that aims to identify the manufacturer of pacemakers and defibrillators from a chest X-Ray image. The app utilizes machine learning, including autonomous learning software leading the algorithm to improve over time. Our study aims to investigate the accuracy of the Pacemaker ID app at detecting the manufacturer of cardiac implantable electronic devices MethodsA total of 200 consecutive x-rays were collected from a two year period at The George Washington University in Washington DC and Centro Medico Hospital in Honduras. Chest X-Rays with implanted devices were scanned using the app. The device manufacturer was recorded from the initial operative report ResultsThe dataset included 146 Medtronic, 23 Boston Scientific, 24 St. Jude and 7 Biotronik devices. The app has a weighted average precision of 85.5%, recall of 73% and F1-score of 77.28%. Dataset accuracy 73%. Medtronic had the most precise prediction at 95.5%, while Biotronik only had 15.2% precision. Boston Scientific had the best recall at 87%. Medtronic, the most abundant in the dataset, showed a high recall of 72% and a very high precision of 95.5%. The F1-score, where Medtronic score the highest with 82%. Biotronik has a low F1-score, at 25% due to the low precision of estimates, particularly because 17% of the Medtronic images were wrongly identified as Biotronik ConclusionsThe PacemakerID application was found to have a 73 % overall accuracy in identifying devices. It has the potential to expedite the interrogation process in hospital settings and improve the quality of care among patients with cardiac devices. Taking into account the following limitationsdevice model, camera type and image quality