**Authors and Affiliations**
Kathryn E. Mangold PhD1, Rickey E. Carter PhD2, Konstantinos C. Siontis MD1, Peter A. Noseworthy MD MBA1, Francisco Lopez-Jimenez MD MBA1, Samuel J. Asirvatham MD1, Paul A. Friedman MD1, Zachi I. Attia, PhD1
1Department of Cardiology, Mayo Clinic, Rochester, MN 55905 USA
2Department of Quantitative Health Sciences, Mayo Clinic, Jacksonville, FL, 32224

**Background**
Smartphones and watches can now record single-lead ECGs, making wearables a potential screening tool for wellness monitoring outside of healthcare settings. Because friends and family often share their smart devices, confirmation that a sample is from a given patient is important before it is added to the electronic health record.

**Methods**
We used 971,337 ECGs from 97,388 distinct patients split into training, validation, and testing sets by patient.  ECGs were then paired so that each pair consists of two distinct ECGs: both from the same patient or from different patients. ECG pairs was first fed into same (Siamese) residual neural network to extract features. These features were then compared, and their differences were passed to a final classifier to yield a similarity score.

**Results**
When using similarity scores to discriminate whether distinct ECGs of a pair came from the same patient or different patients, inputs of single lead and 12-lead medians produced an area under the curve (AUC) of 0.91 and 0.93, respectively. Inputs of single lead and 12-lead short rhythm ECGs produced an AUC of 0.92, and 0.95 respectively.

**Conclusion**
The similar performance of the median and short rhythm ECGs in both single lead and 12-lead configurations underscores the potential use of mobile devices to monitor cardiac health. Because biological aging affects the ECG, when examining pairs within patients’ set of ECGs, the similarity score tends to decrease with large differences in acquisition times. Unexpected rates of change may be key to implementing true personalized medicine by prompting medical evaluation.



**Authors and Affiliations**
Kathryn E. Mangold PhD^1^, Rickey E. Carter PhD^2^, Konstantinos C. Siontis MD^1,^ Peter A. Noseworthy MD MBA^1^, Francisco Lopez-Jimenez MD MBA^1^, Samuel J. Asirvatham MD^1^, Paul A. Friedman MD^1^, Zachi I. Attia, PhD^1^
^1^Department of Cardiology, Mayo Clinic, Rochester, MN 55905 USA
^2^Department of Quantitative Health Sciences, Mayo Clinic, Jacksonville, FL, 32224

**Background**
Smartphones and watches can now record single-lead ECGs, making wearables a potential screening tool for wellness monitoring outside of healthcare settings. Because friends and family often share their smart devices, confirmation that a sample is from a given patient is important before it is added to the electronic health record.

**Methods**
We used 971,337 ECGs from 97,388 distinct patients split into training, validation, and testing sets by patient.  ECGs were then paired so that each pair consists of two distinct ECGs: both from the same patient or from different patients. ECG pairs was first fed into same (Siamese) residual neural network to extract features. These features were then compared, and their differences were passed to a final classifier to yield a similarity score.

**Results**
When using similarity scores to discriminate whether distinct ECGs of a pair came from the same patient or different patients, inputs of single lead and 12-lead medians produced an area under the curve (AUC) of 0.91 and 0.93, respectively. Inputs of single lead and 12-lead short rhythm ECGs produced an AUC of 0.92, and 0.95 respectively.

**Conclusion**
The similar performance of the median and short rhythm ECGs in both single lead and 12-lead configurations underscores the potential use of mobile devices to monitor cardiac health. Because biological aging affects the ECG, when examining pairs within patients’ set of ECGs, the similarity score tends to decrease with large differences in acquisition times. Unexpected rates of change may be key to implementing true personalized medicine by prompting medical evaluation.



Unlocking the Potential of AI in ECG Biometrics: Age-Related Changes, Anomaly Detection, and Data Authenticity in Mobile Health Platforms