RT Journal Article SR Electronic T1 Conditional voxel-wise partial volume correction for emission tomography: Combining a wavelet-based hidden Markov model with a mutual multi-resolution analysis approach JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP 62P OP 62P VO 49 IS supplement 1 A1 Adrien Le Pogam A1 Mathieu Hatt A1 Nicolas Boussion A1 Denis Guilloteau A1 Jean-Louis Baulieu A1 Caroline Prunier A1 Frederico Turkheimer A1 Dimitris Visvikis YR 2008 UL http://jnm.snmjournals.org/content/49/supplement_1/62P.2.abstract AB 247 Objectives: A limitation of the previously proposed voxel-wise mutual multi-resolution analysis (MMA) partial volume correction (PVC) is the use of a global linear model between the two images which could potentially lead to an over correction. The aim of the present study was therefore to develop a methodology for a conditional and improved MMA PVC algorithm. Methods: In order to compare the functional and anatomical images in the wavelet domain, hierarchical tree structures with probabilistic and statistical relationships for the consecutive wavelet decompositions are defined. Hidden Markov Models are then used on the previous tree-structured graph to model the joint statistics of the wavelet coefficients. Based on this modeling a local dissimilarity comparison is performed to localize and extract the main differences creating “conditional correction maps” (CCMs). These CCMs are subsequently used in the original MMA PVC approach keeping or rejecting details of the correction on a local basis. Results: The new algorithm was successfully tested on synthetic and clinical datasets providing the same level of quantitative accuracy in the structures of interest with the original MMA algorithm, without however introducing structures from the anatomical images which are not present in the functional images. Conclusions: A wavelet hidden Markov model has been integrated with the mutual multi-resolution analysis PVC algorithm improving its overall qualitative and quantitative accuracy.