Automatic blood pool identification in contrast ultrasound using principal component analysis

Several cardiovascular parameters of clinical interest can be assessed by indicator dilution techniques. Ultrasound contrast agents have been proposed as non-invasive indicator, showing promising results for blood volume estimation. However, the definition of an optimal region of interest for quantification of the indicator remains a critical step in the procedure, usually performed manually. In this work we present an automatic method to extract indicator dilution curves. Dimensionality reduction is achieved by principal component analysis followed by clustering to identify the different regions of interest. The method is evaluated on in vitro and in vivo datasets, compared to manually defined regions. The average difference was −3.47% ± 3.58% for in vitro volume estimates and the error was 1.29% ± 2.54% for trans-pulmonary mean transit time estimation. The new method allows kinetic parameter estimates in close agreement with those obtained manually; therefore it is a promising alternative for indicator dilution curve extraction.