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

Area of Science:

• Medical Imaging
• Signal Processing
• Radiopharmaceuticals

Background:

• Independent Component Analysis (ICA) is a common technique for signal separation.
• Low signal-to-noise ratio (SNR) in data can impede accurate component estimation.
• Arterial blood sampling in Positron Emission Tomography (PET) can be invasive.

Purpose of the Study:

• To evaluate a novel spatial ICA-based method for extracting blood-related components from PET images.
• To compare the performance of a proposed cost function against kurtosis for PET data analysis.
• To determine the optimal cost function for ICA in PET imaging based on SNR.

Main Methods:

• Spatial ICA was applied to real PET images acquired using three different radiopharmaceuticals.
• A proposed cost function, designed considering component statistical properties, was tested.
• The proposed cost function was compared with kurtosis, a conventional ICA cost function.

Main Results:

• The proposed cost function demonstrated sensitivity to outliers, leading to estimation failures in low SNR PET data.
• Kurtosis, a conventional cost function, showed better performance in certain low SNR scenarios.
• The study highlights the critical role of SNR in selecting an appropriate ICA cost function.

Conclusions:

• The choice of cost function for spatial ICA in PET imaging is highly dependent on the data's signal-to-noise ratio (SNR).
• The proposed cost function is not robust to low SNR conditions and may fail in such scenarios.
• Selecting the appropriate cost function is crucial for reliable component estimation in PET image analysis.