Fig 1.
Scheme for comparison of subjects between S1 and S2 sessions.
Content of the training and test datasets considering all pairwise ID combinations (S2 vs. S1).
Fig 2.
Example of 12-lead average beat patterns from three different subjects (with identity named IDx, IDy, IDz), which are aligned by maximal cross-correlation in lead I to a reference pattern.
The vertical red lines encompass the synchronously extracted 12-lead QRS pattern in a window [-30ms; 70ms] around the R-peak of the reference pattern.
Fig 3.
Example of 12-lead QRS pattern matching between S1 and S2 sessions, taking recordings from equal ID subjects (left panel) and different ID subjects (right panel). (A) The grey approximation span around each QRS pattern (white trace) represents the ones in the corresponding 2D binary matrix binQRS(100x80). (B) The green zones represent the ones in the binary AND matrix for computation of tEQU, matching the time equivalence between the two patterns (black traces). (C) The red elements represent the ones in the binary NAND matrix for computation of aDIF, matching the area difference between the two patterns (black traces).
Table 1.
Median value (quartile range) of tEQU and aDIF features for 12 ECG leads (S1 File).
Statistically different distributions of 460 equal (IDS1 = IDS2) vs. 211140 different (IDS1≠IDS2) identity pairs are found in all leads (p<0.001).
Fig 4.
Training and test ROC curves of single and multi-lead ECG sets.
The line EER (TAR = TRR) illustrates the choice of the operating point on the training ROC.
Table 2.
Human verification performance of single and multi-lead ECG sets: AUC of the training and test ROC.
The bolded values highlight the maximal AUC of the test-ROC for single limb leads, single chest leads, and the multi-lead sets.
Table 3.
Human verification performance of single and multi-lead ECG sets for the EER operating point on the training ROC (Train-TAR = Train-TRR = Train-TVR).
The observed performance on the independent test set has a slight bias Test-TAR>Test-TRR. The bolded values highlight the maximal TVR on the test set for single limb leads, single chest leads, and the multi-lead sets.
Fig 5.
Test-TVR of single and multi-lead ECG sets.
Single leads are ordered according to their spatial neighborhood, i.e. limb leads are presented in ascending order of their spatial angle in the frontal plane (given in brackets, from -30° to 120°); chest leads V1-V6 are presented according to their standard order in the horizontal plane.
Fig 6.
Performance of 12-lead LDA model in function of the number of subjects in the test database.
TAR, TRR and TVR are reported as mean value (min-max range) after test of all possible combinations of 10, 50, 100, 150, 200, 230 subjects within the total test database with 230 subjects. The differences between groups are not statistically significant (p>0.05).
Fig 7.
Gender-specific TVR performance of single and multi-lead LDA models, evaluated for 106 males and 124 females in the test database.
For all leads, TVR (males vs. females) is not statistically significant (p>0.05).
Fig 8.
Age-specific performance of 12-lead LDA model, evaluated for 230 subjects in the test database, divided into six age groups.
The differences between groups are not statistically significant (p>0.05).
Fig 9.
HR-specific performance of 12-lead LDA model, evaluated for 230 subjects in the test database, divided into: (A) 5 groups based on the absolute HR value in S1 session; (B) 3 groups based on the absolute HR change between S1 and S2 sessions (ΔHR). The differences between groups are not statistically significant (p>0.05), except TAR for ≥90 bpm (*p = 0.012).
Table 4.
Verification accuracy reported in published ECG biometric studies, which use at least two recording sessions per subject (distanced from days to years).
Various accuracy metrics reported in other studies (EER, FAR, FRR, TAR, TRR) are transformed to the common metric TVR, using the direct conversions: TVR = 100-EER, TVR = (TAR+TRR)/2, TVR = 100-(FAR+FRR)/2.