Bone recognition in UTE MR images by artificial neural networks for attenuation correction of brain imaging in MR/PET scanners

• Autor(es): Ribeiro, André Filipe dos Santos
• Data: 2012
• Identificador Persistente: http://hdl.handle.net/10451/9257
• Origem: Repositório da Universidade de Lisboa
• Assunto(s): Teses de mestrado - 2012
• Tipo de Documento: Dissertação de mestrado
• Idioma: Inglês
• Orientador(es): Almeida, Pedro Miguel Dinis de, 1968-; Kops, Rota
• Contribuidor(es): Repositório da Universidade de Lisboa

Descrição

Tese de mestrado em Engenharia Biomédica e Biofísica (Radiações em Diagnóstico e Terapia), apresentada à Universidade de Lisboa, através da Faculdade de Ciências, 2012

Aim: Due to space and technical limitations in PET/MR scanners one of the difficulties is the generation of an attenuation correction (AC) map to correct the PET image data. Different methods have been suggested that make use of the images acquired with an ultrashort echo time (UTE) sequence. However, in most of them precise thresholds need to be defined and these may depend on the sequence parameters. In this thesis different algorithm based on artificial neural networks (ANN) are presented requiring little to any user interaction. Material and methods: An MR UTE sequence delivering two images with 0.07 ms and 2.46 ms echo times was acquired from a 3T MR-BrainPET for 9 patients. To correct for intensity inhomogeneities prior to attenuation map estimation a method based on multispetral images was developed and used to correct both images from UTE sequence. The training samples from the corrected images were feed to the proposed algorithms for learning and the methods posterior used for classification. The generated AC maps were compared to co-registered CT images based on the co-classification voxels, dice coefficients and sensitivity correction map (for the 9 patients), and relative differences (for 4 patients) in reconstructed PET images. Results: In overall the methods proposed showed high dice coefficients for air and soft tissue and lower to bone. Adittionaly, the proposed methods showed to present higher dice coefficients than remain methods. High linear correlation between the sensitivity correction maps was verified for all methods. The reconstructed PET images showed mean relative differences 5% for all methods except keereman method, where a mean of 6% was observed. Discussion: The different analysis showed slightly different results regarding the methods that perform best. Nevertheless, all the analysis showed that the methods developed work similar to better than the ones curently proposed. Conclusion: The methods aided by the template image showed to be more robust and with higher specificity than the ones without, altough loosing in sensitivity. Finally, the continuous methods developed showed to be promising as they can estimate different attenuation coefficients within a certain range for the same tissue and therefore account for different densities.