ECG Authentication Based on Non-Linear Normalization under Various Physiological Conditions

Q: What are the key findings of this study?

Non-linear ECG beat normalization achieved 99.05% authentication accuracy in the resting state and 88.14% in the non-resting state.

Q: What is the clinical evidence from this study?

Population: Healthy adults without heart disease undergoing ECG-based biometric authentication (n=15). Intervention: Non-linear normalization of ECG beats for biometric authentication vs. Existing ECG normalization methods and non-normalized ECG beats. Primary outcome: Authentication accuracy (ACC) in resting and non-resting states using Support Vector Machine with Gaussian kernel (SVM-G).

Q: What does this research mean for the field?

Non-linear normalization of ECG beats improves waveform similarity under heart rate fluctuations, achieving 99.05% biometric authentication accuracy in the resting state and 88.14% in non-resting states. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Ho Bin Hwang; Hyeokchan Kwon; Byungho Chung; Jongshill Lee; In Young Kim

doi:10.3390/s21216966

October 20, 2021SensorsOpen Access

ECG Authentication Based on Non-Linear Normalization under Various Physiological Conditions

View Full Paper

Ask AI

Bookmark

Share

Key result

Non-linear ECG normalization achieves ~99% resting authentication accuracy.

n=15

Figures

Why the study?

ECG morphological changes driven by physical and psychological factors can impede biometric authentication under fluctuating heart rates.

Does non-linear normalization of ECG beats improve biometric authentication accuracy and waveform similarity across various physiological conditions in healthy adults?

Key Points

To develop and evaluate a non-linear normalization method for electrocardiogram (ECG) waveforms that preserves biometric authentication performance across varying heart rates caused by physical and emotional activities.
Extracted ECG beats under resting and non-resting conditions (including listening to calm/exciting music, watching relaxed/scary videos, and exercising) and applied section-by-section non-linear normalization based on heart rate analysis.
Evaluated outlier beat rejection against four established normalization methods using independent t-tests and Mann-Whitney tests.
Trained classification models on resting-state data and tested authentication performance in Gaussian Support Vector Machines (SVM-G) using accuracy (ACC), False Acceptance Rate (FAR), False Rejection Rate (FRR), and Equal Error Rate (EER).
The proposed normalization method significantly reduced outlier beat discarding compared to existing methods (t-test p < 0.001 and Mann-Whitney p < 0.01) while increasing average beat similarity by 23.7% in resting states and 43% in non-resting states relative to unnormalized beats.
In the resting state, the SVM-G authentication model achieved 99.05% ACC, 0.85% FAR, 1.04% FRR, and 0.77% EER.
Across all non-resting conditions, the SVM-G model attained 88.14% ACC, 2.6% FAR, 21% FRR, and 9.83% EER, achieving 93.55% accuracy during 5–6% heart rate changes and 90.87% accuracy during 7–8% heart rate changes, though performance deteriorated during exercise with a 48% heart rate change.

Structured PICO

Does non-linear normalization of ECG beats improve biometric authentication accuracy and waveform similarity across various physiological conditions in healthy adults?

P

Population

15 healthy adults without heart disease (mean age 27.29 years, 13 males and 2 females) undergoing ECG measurements across various physiological conditions over 2 to 11 months.

I

Intervention

Non-linear time-axis normalization of ECG beats (segment-by-segment resampling of PR interval, QRS complex, ST interval, and TP segment to standard duration at 70 bpm with z-score amplitude normalization) combined with 29 fiducial features and a Gaussian kernel Support Vector Machine (SVM-G) classifier.

C

Comparator

Conventional linear/partial ECG normalization methods (Arteaga-Falconi et al., Fatemian et al., Choi et al., Nobunaga et al.), alternative classifiers (SVM-Q, kNN, Decision Tree), and unnormalized ECG beats.

O

Outcome

Biometric authentication performance (Accuracy [ACC], False Acceptance Rate [FAR], False Rejection Rate [FRR], and Equal Error Rate [EER]) and waveform similarity (Pearson cross-correlation coefficient) in resting and non-resting states.surrogate

Limitations

Small sample size (15 subjects) compared to other studies
Authentication performance degrades severely when heart rate change is 10% or more (such as during exercise) due to uncorrected amplitude differences
Difficult to determine precisely when authentication performance is reduced due to sharp changes in heart rate
Experimented with fewer subjects (n=15) compared to other studies
Authentication performance deteriorates significantly during exercise or when heart rate changes by 10% or more due to amplitude variations

Implication

Non-linear segment-by-segment time normalization of ECG waveforms improves biometric user authentication accuracy across mild-to-moderate physiological heart rate fluctuations, though further amplitude normalization is needed for high-heart-rate exercise conditions.

Abstract

The development and use of wearable devices require high levels of security and have sparked interest in biometric authentication research. Among the available approaches, electrocardiogram (ECG) technology is attracting attention because of its strengths in spoofing. However, morphological changes of ECG, which are affected by physical and psychological factors, can make authentication difficult. In this paper, we propose authentication using non-linear normalization of ECG beats that is robust to changes in ECG waveforms according to heart rate fluctuations in various daily activities. We performed a non-linear normalization method through the analysis of ECG alongside heart rate, evaluating similarities and authenticating the performance of our new method compared to existing methods. Compared with beats before normalization, the average similarity of the proposed method increased 23.7% in the resting state and 43% in the non-resting state. After learning in the resting state, authentication performance reached 99.05% accuracy for the resting state and 88.14% for the non-resting state. The proposed method can be applicable to an ECG-based authentication system under various physiological conditions.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Hwang et al. (2021) studied Healthy adults without heart disease undergoing ECG-based biometric authentication (n=15). Non-linear normalization of ECG beats for biometric authentication vs. Existing ECG normalization methods and non-normalized ECG beats was evaluated on Authentication accuracy (ACC) in resting and non-resting states using Support Vector Machine with Gaussian kernel (SVM-G). Non-linear ECG beat normalization achieved 99.05% authentication accuracy in the resting state and 88.14% in the non-resting state.

synapsesocial.com/papers/6a74a162f5fc394144032dfb https://doi.org/10.3390/s21216966

View Full Paper

Ask AI

Bookmark

Share

ECG Authentication Based on Non-Linear Normalization under Various Physiological Conditions

Figures

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

ECG Authentication Based on Non-Linear Normalization under Various Physiological Conditions

Figures

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading