Neuroimaging analysis aims to support clinicians in the diagnosis of neurological diseases by using radiological images. Positron emission tomography (PET) is a nuclear medicine imaging technique used to produce three-dimensional images of the human brain for neurological studies. Due to the large number of generated images, there is a lot of effort in defining computer based tools to analyze and classify brain images. Such analyses are used to identify cerebral regions of interest (ROI) related to specific neurodegenerative diseases. Statistical tools, such as SPM (for Statistical Parametric Mapping) and its MarsBar plugin, are largely used by physicians for ROIs identification and for image analysis. Nevertheless, large datasets analysis (e.g. studying pathologies for many patients and for large sets of PET images) requires repetitive SPM procedures for each patient’s image, mainly due to the lack of (i) automatic procedures for analysing set of patients, and (ii) validation of using SPM versus patient magnetic resonance as reference brain templates. Finally, SPM analysis requires human intervention, and there is no automatic system guiding physicians for pathologies identification. As a contribution for the latter issue, we defined an automatic classification tool using topological relations among ROIs to support physicians while studying a new patient. Starting from a set of known pathologies associated to medical annotated PET images (i.e. associated to neurological pathologies), we used SPM and MarsBaR tools to define a reference PET images dataset; ROIs extracted from input PET images have been compared with known dataset and classified, suggesting physicians with (a subset of) pathologies associated to those PET images. Experiments showed that the classifier performs well. Moreover, in order to improve the repeatability of experiments with large datasets, we use an SPM plugin called AutoSPET, which allows to perform SPM analysis on a large PET images dataset, using different SPM plugins within a unified user interface, and allowing to simply run statistical analyses. AutoSPET is available on our server and also as an SPM plugin on the SPM website. Finally we report experiments to validate the use of the standard T1 SPM template versus the magnetic resonance ones.
Methodologies for the analysis and classification of PET neuroimages
Vizza, P.;Tradigo, G.;
2013-01-01
Abstract
Neuroimaging analysis aims to support clinicians in the diagnosis of neurological diseases by using radiological images. Positron emission tomography (PET) is a nuclear medicine imaging technique used to produce three-dimensional images of the human brain for neurological studies. Due to the large number of generated images, there is a lot of effort in defining computer based tools to analyze and classify brain images. Such analyses are used to identify cerebral regions of interest (ROI) related to specific neurodegenerative diseases. Statistical tools, such as SPM (for Statistical Parametric Mapping) and its MarsBar plugin, are largely used by physicians for ROIs identification and for image analysis. Nevertheless, large datasets analysis (e.g. studying pathologies for many patients and for large sets of PET images) requires repetitive SPM procedures for each patient’s image, mainly due to the lack of (i) automatic procedures for analysing set of patients, and (ii) validation of using SPM versus patient magnetic resonance as reference brain templates. Finally, SPM analysis requires human intervention, and there is no automatic system guiding physicians for pathologies identification. As a contribution for the latter issue, we defined an automatic classification tool using topological relations among ROIs to support physicians while studying a new patient. Starting from a set of known pathologies associated to medical annotated PET images (i.e. associated to neurological pathologies), we used SPM and MarsBaR tools to define a reference PET images dataset; ROIs extracted from input PET images have been compared with known dataset and classified, suggesting physicians with (a subset of) pathologies associated to those PET images. Experiments showed that the classifier performs well. Moreover, in order to improve the repeatability of experiments with large datasets, we use an SPM plugin called AutoSPET, which allows to perform SPM analysis on a large PET images dataset, using different SPM plugins within a unified user interface, and allowing to simply run statistical analyses. AutoSPET is available on our server and also as an SPM plugin on the SPM website. Finally we report experiments to validate the use of the standard T1 SPM template versus the magnetic resonance ones.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
