RESUMO

BACKGROUND: Multiple sclerosis (MS) is a neurologic disease of the central nervous system which affects almost three million people worldwide. MS is characterized by a demyelination process that leads to brain lesions, allowing these affected areas to be visualized with magnetic resonance imaging (MRI). Deep learning techniques, especially computational algorithms based on convolutional neural networks (CNNs), have become a frequently used algorithm that performs feature self-learning and enables segmentation of structures in the image useful for quantitative analysis of MRIs, including quantitative analysis of MS. To obtain quantitative information about lesion volume, it is important to perform proper image preprocessing and accurate segmentation. Therefore, we propose a method for volumetric quantification of lesions on MRIs of MS patients using automatic segmentation of the brain and lesions by two CNNs. METHODS: We used CNNs at two different moments: the first to perform brain extraction, and the second for lesion segmentation. This study includes four independent MRI datasets: one for training the brain segmentation models, two for training the lesion segmentation model, and one for testing. RESULTS: The proposed brain detection architecture using binary cross-entropy as the loss function achieved a 0.9786 Dice coefficient, 0.9969 accuracy, 0.9851 precision, 0.9851 sensitivity, and 0.9985 specificity. In the second proposed framework for brain lesion segmentation, we obtained a 0.8893 Dice coefficient, 0.9996 accuracy, 0.9376 precision, 0.8609 sensitivity, and 0.9999 specificity. After quantifying the lesion volume of all patients from the test group using our proposed method, we obtained a mean value of 17,582 mm3. CONCLUSIONS: We concluded that the proposed algorithm achieved accurate lesion detection and segmentation with reproducibility corresponding to state-of-the-art software tools and manual segmentation. We believe that this quantification method can add value to treatment monitoring and routine clinical evaluation of MS patients.

RESUMO

A Medicina Nuclear, como especialidade de obtenção de imagens médicas é um dos principais procedimentos utilizados hoje nos centros de saúde, tendo como grande vantagem a capacidade de analisar o comportamento metabólico do paciente. Este projeto está baseado em imagens médicas obtidas através da modalidade PET (Positron Emission Tomography). Para isso, foi desenvolvida uma estrutura de processamento de imagens tridimensionais PET, constituída por etapas sucessivas que se iniciam com a obtenção das imagens padrões (gold standard), sendo utilizados para este fim volumes simulados do Ventrículo Esquerdo do Coração criadas como parte do projeto, assim como phantoms gerados com o software NCAT-4D. A seguir, nos volumes simulados é introduzido ruído Poisson que é o ruído característico das imagens PET. Na sequência é executada uma etapa de pré-processamento, utilizando alguns filtros 3D tais como o filtro da mediana, o filtro da Gaussiana ponderada e o filtro Anscombe/Wiener. Posteriormente é aplicada a etapa de segmentação, processo baseado na teoria de Conectividade Fuzzy sendo implementadas quatro diferentes abordagens 3D: Algoritmo Genérico, LIFO, kTetaFOEMS e Pesos Dinâmicos. Finalmente, um procedimento de avaliação conformado por três parâmetros (Verdadeiro Positivo, Falso Positivo e Máxima Distância) foi utilizado para mensurar o nível de eficiência e precisão do processo. Constatou-se que o par Filtro - Segmentador constituído pelo filtro Anscombe/Wiener junto com o segmentador Fuzzy baseado em Pesos Dinâmicos proporcionou os melhores resultados, com taxas de VP e FP na ordem de 98,49 ± 0,27% e 2,19 ± 0,19%, respectivamente, para o caso do volume do Ventrículo Esquerdo simulado. Com o conjunto de escolhas feitas ao longo da estrutura de processamento, encerrou-se o projeto analisando um número reduzido de volumes pertencentes a um exame PET real, obtendo-se a quantificação dos volumes.

The Nuclear medicine, as a specialty to obtain medical images is very important, and it has became one of the main procedures utilized in Health Care Centers to analyze the metabolic behavior of the patient. This project was based on medical images obtained by the PET modality (Positron Emission Tomography). Thus, we developed a framework for processing Nuclear Medicine three-dimensional images of the PET modality, which is composed of consecutive steps that start with the generation of standard images (gold standard) by using simulated images of the Left Ventricular Heart, such as phantoms obtained from the NCAT-4D software. Then, Poisson quantum noise was introduced into the whole volume to simulate the characteristic noises in PET images. Subsequently, the pre-processing step was executed by using specific 3D filters, such as the median filter, the weighted Gaussian filter, and the Anscombe/Wiener filter. Then the segmentation process, which is based on the Fuzzy Connectedness theory, was implemented. For that purpose four different 3D approaches were implemented: Generic, LIFO, kTetaFOEMS, and Dynamic Weight algorithm. Finally, an assessment procedure was used as a measurement tool to quantify three parameters (True Positive, False Positive and Maximum Distance) that determined the level of efficiency and precision of our process. It was found that the pair filter - segmenter formed by the Anscombe/Wiener filter together with the Fuzzy segmenter based on Dynamic Weights provided the best results, with VP and FP rates of 98.49 ± 0.27% and 2.19 ± 0.19%, respectively, for the simulation of the Left Ventricular volume. Along with the set of choices made during the processing structure, the project was finished with the analysis of a small number of volumes that belonged to a real PET test, thus the quantification of the volumes was obtained.