Image similarity-based cardiac rhythm device identification from X-rays using feature point matching.

Area of Science:

• Medical Imaging
• Computer Vision
• Cardiology

Background:

• Accurate identification of cardiac implantable electronic devices (CIEDs) is critical in emergency situations.
• Existing artificial neural network models for CIED recognition require extensive medical data.
• A novel, data-efficient method is needed for CIED identification from X-ray images.

Purpose of the Study:

• To develop and evaluate a novel method for identifying CIEDs from chest X-ray images using feature point matching.
• To retrieve identical CIED images from a database without requiring large training datasets.

Main Methods:

• A dataset of 653 X-ray images from 456 patients was utilized.
• Scale-Invariant Feature Transform (SIFT) algorithm extracted keypoints from images.
• Brute-force matching identified paired feature points, and average Euclidean distance determined image similarity.

Main Results:

• Manufacturer classification achieved an average accuracy of 97.0%, precision of 0.97, recall of 0.96, and F1-score of 0.96.
• Model classification achieved an average accuracy of 93.2%, precision of 0.94, recall of 0.92, and F1-score of 0.93.
• Performance metrics surpassed those of previously reported machine learning models.

Conclusions:

• Feature point matching offers a viable and efficient alternative for CIED identification from X-ray images.
• This method reduces the dependency on large-scale medical datasets for CIED classification.
• The proposed technique enhances diagnostic capabilities in emergent clinical scenarios.