Cardiovascular Implantable Electronic Devices -> Diagnostic Devices & Sensors: -> Device Technology D-PO03 - Poster Session III (ID 48) Poster

D-PO03-088 - Machine Versus Machine: A Head-to-head Comparison Of Machine Learning Algorithms For Identification Of Cardiac Devices On Chest Radiography (ID 1100)


Background: Machine learning techniques are becoming more commonly used in healthcare. Two recently published machine learning algorithms for identification of implanted cardiac devices have been described: Pacemaker-ID (PID) and Pacemaker Identification with Neural Networks (PPMnn). PID is available as a mobile phone app (PIDa) and web version (PIDw) whereas PPMnn is available only as web version.
Objective: To compare accuracy of PID and PPMnn machine learning algorithms in identification of device manufacturer in a real-world setting.
Methods: Chest X-rays from patients with pacemakers and defibrillators from four device manufacturers were retrospectively obtained from a single institution. Cropped images of the device generator were analyzed using each algorithm as well as the Cardia-X visual flowchart algorithm. Accuracy of each method in determining device manufacturer was compared.
Results: Chest X-rays from 500 individuals were included. PIDa achieved highest overall accuracy for all devices at 91%. PPMnn was most accurate in identification of pacemakers and PIDa was most accurate in identification of defibrillators.
Conclusion: Machine learning algorithms can achieve similar or higher accuracy in identification of cardiac device manufacturers using chest radiography. The PacemakerID mobile phone app (PIDa) overall performed best. Web versions had difficulty identifying Biotronik devices.