Practical Consideration about Cost Functions of Spatial Independent Component Analysis in Medical Image Processing

Independent component analysis (ICA) is widely used for signal separation in various fields. However, low signal to noise ratio (SNR) of data sometimes causes failures in component estimation. We have proposed a spatial ICA-based method for extracting a blood-related component from medical images measured with positron emission tomography (PET) to omit arterial blood sampling, in which a cost function was designed in consideration of statistical properties of components. In this study, spatial ICA with the proposed cost function or kurtosis, a conventional cost function, was applied to real PET images measured with three kinds of radiopharmaceuticals, and the estimation results were compared. The proposed cost function fails to estimate components from the PET data with low SNR because it is sensitive to outliers. Experimental results suggest that a cost function should be selected depending on SNR of data.