Comparison of Machine Learning Algorithms for Identification of Implantable Cardiac Devices on Chest Radiography

Chudow, Jay; Jones, Davis; Weinreich, Michael; Zaremski, Lynn; Lee, Suegene; Weinreich, Brian; Fisher, John D.; Krumerman, Andrew; Ferrick, Kevin J.

doi:10.1016/s0735-1097(20)34150-4

Cited by 2 publications

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The sample size was calculated based on the results of Jay Chudow (10), using the formula proposed by Lachenbruch in 1992 (11), implemented in the biostatUZH R package by the sampleSizeMcNemar function. For an estimated overall accuracy of 80%, an alpha of 0.05, the minimum number of images was 395.…”

Section: Discussionmentioning

confidence: 99%

The accuracy of two algorithms of artificial intelligence based on neural networks and the CaRDIA-X algorithm in the identification of electronic implantable cardiac devices by chest x-rays.

Córdoba

Rodriguez

García

et al. 2022

Preprint

View full text Add to dashboard Cite

Objectives: In this study, we aim to describe the diagnostic accuracy of two applications neural networks-based system and a visual algorithm performed by different evaluators to identify the manufacturer of electronic implantable cardiac devices by chest x-rays. Background: cardiac rhythm devices frequently require interrogation, and they have different software depending on the manufacturer. Currently, there are a visual algorithm and two applications based on artificial intelligence for the identification of the manufacturer from chest radiographs. Methods: Retrospective trial between January 2010 and December 2021 at a single institution. Chest radiographs were obtained from patients with cardiac devices; they were cropped and resized to 224 by 224 pixels. Then, they were analyzed using the applications Pacemaker ID with a cell phone, Pacemaker ID web and PPMnn web, and the visual algorithm CaRDIA-X performed by evaluators at different levels of training. Results: 400 radiographic images with cardiac devices were collected comprising 4 manufacturers and 40 different models. The agreement for Pacemaker ID with a cell phone was 90.6% ( p <0.001), for Pacemaker ID web was 81.2% ( p < 0.001); and for PPMnn web was 82% ( p < 0.001). The agreement from the CaRDIA-X algorithm performed by 4 evaluators ranged from 73.8% to 97.7% ( p < 0.001). Conclusions: The use of applications based on neural networks offers a good agreement in the identification of the manufacturer and is a tool for clinical use. In our paper, the visual algorithm has a better agreement in identifying the manufacturer and it doesn’t require much training.

show abstract

Section: Discussionmentioning

confidence: 99%