On the measurement of physiological similarity between independent components: Time-structure versus frequency-based methods

This work explored two methodologies for clustering independent components (ICs) into physiological groups corresponding to maternal respiratory (MR), maternal cardiac (MC), foetal cardiac (FC), or noisy (N) activities. The methods, based on frequency content (S) or time-structure (R) analyses, were tested on 750 ICs (extracted from 25 abdominal phonograms by using single-channel independent component analysis, SCICA). Results showed that the S-based methodology is more reliable at clustering similar ICs than the R-based method. On the other hand, the R-based methodology not only clusters ICs, but also identifies their physiological origin, which is a desired quality. These characteristics make both schemes interesting for automatic and fast classification of ICs extracted from the abdominal phonogram. These results were so promising that current work has already combined both methods into an enhanced version for grouping physiological ICs extracted by SCICA. Future work will analyse foetal traces for well-being information recovery.