Table 2 AF prediction performance at a given sensitivity of 0.80.

From: Prediction of atrial fibrillation from at-home single-lead ECG signals without arrhythmias

Feature set

Age group

Input ECG length

Tot. recordings

Prevalence (%)

Sensitivity

Specificity

Precision

F1 score

AG

<65

1 day

11,820

2.2

0.80

0.58

0.04

0.08

AG

≥65

1 day

11,329

6.5

0.80

0.26

0.07

0.13

AG

Overall

1 day

23,149

4.3

0.80

0.49

0.07

0.12

AG + HRV

<65

1 day

11,820

2.2

0.80

0.66

0.05

0.09

AG + HRV

≥65

1 day

11,329

6.5

0.80

0.37

0.08

0.15

AG + HRV

Overall

1 day

23,149

4.3

0.80

0.56

0.08

0.14

DL Only

<65

1 day

11,820

2.2

0.80

0.72

0.06

0.11

DL Only

≥65

1 day

11,329

6.5

0.80

0.51

0.10

0.18

DL Only

Overall

1 day

23,149

4.3

0.80

0.65

0.09

0.17

All Features

<65

1 day

11,820

2.2

0.80

0.76

0.07

0.13

All Features

≥65

1 day

11,329

6.5

0.80

0.50

0.10

0.18

All Features

Overall

1 day

23,149

4.3

0.80

0.65

0.09

0.17

  1. Specificity, precision, and F1 score evaluated at a given sensitivity of 0.80 when considering a 1-day single-lead ECG as input, for people under the age of 65, for people above 65, and overall. Prevalence and total number of recordings for each group are also reported. The performance is reported for the model with four different input configurations: age and gender (AG); age, gender, and HRV (AG + HRV); deep learning features only (DL Only); all input available (All Features).
Back to article page