Table 4 Results of machine learning algorithm for detecting AF rhythms in the ECG and IH data.

From: In-ear infrasonic hemodynography with a digital health device for cardiovascular monitoring using the human audiome

 

No. of subjects

No. of segments

Total time (s)

Accuracy

Precision

Recall

ECG

SR: n = 25

606

18,180

1.00

0.99

1.00 [0.99, 1.0]

 

AF: n = 15

458

13,740

 

1.00

0.99 [0.98, 1.0]

IH

   

0.99

0.99

0.99 [0.98, 1.0]

     

0.99

0.99 [0.98, 1.0]

  1. Number of SR and AF subjects, number of 30-second segments of cardiac rhythms and their total duration (common for the ECG and IH data), as well as accuracy, precision, and recall metrics for detecting AF rhythms with a random forest classifier model. The model was trained and validated using external PhysioNet ECG data and then applied to individual ECG (top row) and IH data (bottom row). Recall for the AF and SR samples corresponds to Sensitivity and Specificity in detecting AF, respectively. Numbers in square brackets adjacent to Recall values correspond to confidence intervals (95%) calculated as Wilson score intervals21.
Back to article page