Lesion enhancement in dynamic positron emission tomography using subspace filtering

The high dimensionality and high level noise of dynamic positron emission tomography (PET) data make the identification of the kinetic feature of lesions difficult. Principle component analysis (PCA) is a well-known multivariate image analysis technique to reduce the dimensions and noise level in dynamic PET data. However, the contrast of small lesions is often reduced by PCA, due to the lack of using a priori physiological knowledge of lesions. In this paper, a new method, modified from the generalized sidelobe canceler (GSC), is proposed to incorporate the physiological features extracted from the visible tumor and normal tissue in order to enhance the non-palpable lesions in the background noise. Two different schemes are adopted to implement the modified GSC. One uses both physiological features of lesions and normal tissue, while the other employs only that of lesions. The multidimensional maximum likelihood estimation is applied to find physiological factors which span time activity curve subspaces of lesions and normal tissue from dynamic PET data. Results show that the proposed modified GSC can substantially stretch the contrast of small lesions, and thus is able to aid in detecting non-palpable tumors. The images resulting from post-processing the FBP by the modified GSC are compared with those of the OSEM reconstruction as well as the PCA. The analytical analysis of the method is also described