Dependent Component Analysis - Biomedical Signal Analysis: Contemporary Methods and Applications - page 157

(a) ECG recordings (b) extracted sources

(c) MECG part (d) fetal ECG part

Figure 5.6
Independent subspace analysis with known block structure m =(2 , 1) is applied to
fetal ECG. (a) shows the ECG recordings. The underlying FECG (4 heartbeats) is
partially visible in the dominating MECG (3 heartbeats). (b) gives the extracted
sources using ISA with the Hessian source condition from table 5.1 with 500
Hessian matrices. In (c) and (d) the projections of the mother sources (first two
components from (b)) and the fetal source (third component from (b)) onto the
mixture space (a) are plotted.

Application to ECG data

Finally, we report the example from [249] on how to apply the Hes-
sian ISA algorithm to a real-world data set. Following [43], we show
how to separate fetal ECG (FECG) recordings from the mother's ECG
(MECG). Our goal is to extract an MECG component and an FECG
component; however we cannot expect to find only a one-dimensional
MECG due to the fact that projections of a three-dimensional vector
(electric) field are measured. Hence, modeling the data by a multidi-
mensional BSS problem with k = 2 (but allowing for an additional
one-dimensional component) makes sense. Application of ISA extracts a
two-dimensional MECG component and a one-dimensional FECG com-
ponent. After block permutation we get estimated mixing matrix A and