Cardiac Rhythm Device Identification Using Neural Networks.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Machine Learning for Healthcare
• Radiographic Analysis of Cardiac Devices

Background:

• Medical staff require rapid and precise identification of pacemaker and defibrillator models.
• Current methods rely on manual comparison with flowcharts, which can be time-consuming and error-prone.

Purpose of the Study:

• To develop and validate a neural network-based system for identifying cardiac rhythm device manufacturers and model groups from chest radiographs.
• To compare the AI system's performance against human cardiologists in device identification.

Main Methods:

• A convolutional neural network was trained on 1,451 radiographic images of 45 pacemaker/defibrillator models from 5 manufacturers.
• The network was tested on 225 images, with performance compared to 5 cardiologists using a flowchart.

Main Results:

• The neural network achieved 99.6% accuracy in manufacturer identification and 96.4% in model group identification.
• Human cardiologists had a median manufacturer identification accuracy of 72.0%, with model group identification not possible.
• The AI system significantly outperformed cardiologists in identifying device manufacturers (p < 0.0001).

Conclusions:

• A neural network can accurately identify cardiac rhythm device manufacturers and model groups from radiographs, surpassing human performance.
• This AI system offers a potential to expedite diagnosis and treatment for patients with these devices.
• The developed system is publicly accessible online.