Finding similar ECGs in a large 12-lead ECG database

Automated matching of a patient's 12-lead ECG within a large 12-lead ECG database to find similar ECGs has many potential applications including searching for examples, diagnosing ECG by probability estimation, or confirming patient identification. We created a morphology-similarity matching algorithm and reported the performance in this study. The study set consisted of 24,262 ECGs from 8,663 subjects. Similar ECGs were found by two methods, exhaustive search of pair-wise template matching and fast query using a k-dimensional tree architecture and a processed version ofthe ECG signal as feature vector. Two ECGs were similar if they came from the same patient. For each ECG in the study set, 20 nearest neighbors were extracted from the database. Sensitivities were calculated for finding any and all of the ECGs from the same patient in the set of 20 nearest neighbors. In the exhaustive search, sensitivities were 68% and 37% for finding any and all of the ECGs from the same patient respectively. With the fast query, sensitivities were 48% (any) and 30% (all). Sensitivity for the fast query increased to 56% (any) and 37% (all) when extracting 50 nearest neighbors. We conclude that a low complexity but fast query can be used to find similar 12-lead ECGs from a large database.