Development of a Osirix plug-in for non-gaussian diffusion MRI data: application to Breast

• Autor(a) principal: Amorim, Ana Carolina Costa
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/33790
• topic: Diffusion Kurtosis Imaging; Intravoxel Incoherent Motion; Gamma Distribution; Truncated; Stretched-Exponential; Plug-in; Domínio/Área Científica::Engenharia e Tecnologia::Outras Engenharias e Tecnologias

Breast cancer is the second leading cause of death by cancer and is the second type of cancer that is the most common among women, causing the death of 1500 women, every year, in Portugal. Over the past decades, with the improvement of imaging techniques and therapeutics, breast cancer mortality rate has been decreasing substantially. One of the imaging techniques used is diffusion-weighted imaging (DWI), which is the magnetic resonance imaging method that is the most sensitive in detecting invasive breast cancer. There are several extensions of this technique that can be based on the premise that the environment is not homogeneous (non-Gaussian models). The non-Gaussian models, take into account the presence of barriers and compartments that restrict the water molecules movement in the biological tissues, allowing the distinction between malignant and benign lesions in DWI. Taking into account that diffusion-weighted imaging is a non invasive method, the diagnosis through the application of these models can become an alternative to biopsy. The main goal of this dissertation was the development of a OsiriX plug-in that performs the non-Gaussian diffusion analysis of DWI in magnetic resonance imaging data. Five non-Gaussian diffusion models were implemented in the application: diffusion kurtosis imaging, intravoxel incoherent motion, gamma distribution, truncated and stretched-exponential. Parametric maps and region-of interest (ROI) parametric values were obtained for each model. In order to fit the various non-Gaussian diffusion models, the Levenberg-Marquardt algorithm was used. This algorithm finds coefficients x to best fit the nonlinear function to the data. The x coefficients results in parametric maps that are able to identify the tumour and also distinguish between malignant and benign lesions provided a comparison with the literature. During the application development, these functionalities were also used to study the dependence of fitted parameters on the number of b-values used and on the image noise. An interface was developed with Objective-C on Xcode for OsiriX where the user gets to choose from five diffusion models in order to obtain parameric maps or values. It also informs on the goodness of fit and it provides fitting plots of the data. The references of the models and extra information are also available in an help button. Finally, the application was tested successfully with breast DWI images of benign and malignant tumors.