Conditional partial volume correction for emission tomography: A wavelet-based hidden Markov model and multi-resolution approach

The spatial resolution in emission tomography induces partial volume effects, leading to under-estimation of the real uptake and activity spillover. They can be corrected using either region of interest or voxel based approaches. We developed a voxel wise correction, based on the wavelet transform of two co-registered images in order to insert high resolution details of the anatomical image into the functional one. A limitation is the use of a global model which may create artefacts in the corrected image where there is no correlation between the two modalities. The aim of this work was to develop a conditional correction: a comparison between the two modalities using hidden Markov modelling and multi-resolution analysis was developed to allow such a correction. The process was successfully tested on synthetic and clinical images avoiding artefacts.