A carregar...
Publicação
Inteligência artificial para deteção de tumores cerebrais em imagens médicas
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 61.14 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A aplicação de Inteligência Artificial (IA) na análise de imagens médicas tem transformado a deteção precoce e o diagnóstico de diversas doenças, como tumores, lesões e anomalias em órgãos vitais. Modelos baseados em Machine learning (ML) e Deep learning (DL), especialmente redes neuronais convolucionais (CNN), têm demonstrado elevado desempenho na segmentação, classificação e localização de estruturas patológicas em exames como ressonâncias magnéticas, tomografias e radiografias. Estas abordagens permitem não só uma maior precisão diagnóstica, como também reduzem o tempo de análise e a carga de trabalho dos profissionais de saúde. A integração de técnicas supervisionadas e não supervisionadas possibilita uma extração robusta de características relevantes, mesmo em contextos com dados limitados ou rótulos incompletos. Neste sentido a IA, afirma-se como uma ferramenta promissora no apoio à decisão médica, contribuindo para o avanço da medicina da precisão. O principal objetivo deste trabalho é o desenvolvimento de modelos baseados em IA para a deteção de tumores cerebrais a partir da análise de imagens médicas. Para tal, foram utilizadas diferentes arquiteturas de redes CNN, incluindo VGG-19, ResNet, EfficientNet, entre outras, em combinação com técnicas de Explainable
Artificial Intelligence (XAI), de modo a garantir não só elevados níveis de desempenho, mas também interpretabilidade dos resultados. Os modelos foram treinados e avaliados com recurso a bases de dados públicas de ressonância magnética cerebral, permitindo a classificação de diferentes tipos de tumores. Os resultados
obtidos demonstraram valores elevados de Accuracy, Recall e Area Under the Curve Receiver Operating Characteristic (AUROC), o que evidencia a eficácia da solução proposta na deteção de tumores cerebrais e confirma o seu potencial como sistema de apoio ao diagnóstico clínico.
The application of AI in medical image analysis has transformed the early detection and diagnosis of various diseases, such as tumors, lesions, and anomalies in vital organs. Models based on ML and DL, particularly convolutional neural networks (CNNs), have demonstrated high performance in the segmentation, classification, and localization of pathological structures in examinations such as magnetic resonance imaging (MRI), computed tomography (CT), and X-rays. These approaches not only enable greater diagnostic accuracy but also reduce analysis time and the workload of healthcare professionals. The integration of supervised and unsupervised techniques allows for robust extraction of relevant features, even in contexts with limited data or incomplete labels. In this sense, AI establishes itself as a promising tool for medical decision support, contributing to the advancement of precision medicine. The main objective of this work is the development of AI-based models for the detection of brain tumors through the analysis of medical images. To this end, different CNN architectures were employed, including VGG-19, ResNet, and EfficientNet, among others, in combination with Explainable Artificial Intelligence (XAI) techniques, aiming to ensure not only high levels of performance but also interpretability of the results. The models were trained and evaluated using publicly available brain MRI datasets, enabling the classification of different types of tumors. The results obtained demonstrated high Accuracy, Recall, and AUROC values, evidencing the effectiveness of the proposed solution in detecting brain tumors and confirming its potential as a clinical decision support system.
The application of AI in medical image analysis has transformed the early detection and diagnosis of various diseases, such as tumors, lesions, and anomalies in vital organs. Models based on ML and DL, particularly convolutional neural networks (CNNs), have demonstrated high performance in the segmentation, classification, and localization of pathological structures in examinations such as magnetic resonance imaging (MRI), computed tomography (CT), and X-rays. These approaches not only enable greater diagnostic accuracy but also reduce analysis time and the workload of healthcare professionals. The integration of supervised and unsupervised techniques allows for robust extraction of relevant features, even in contexts with limited data or incomplete labels. In this sense, AI establishes itself as a promising tool for medical decision support, contributing to the advancement of precision medicine. The main objective of this work is the development of AI-based models for the detection of brain tumors through the analysis of medical images. To this end, different CNN architectures were employed, including VGG-19, ResNet, and EfficientNet, among others, in combination with Explainable Artificial Intelligence (XAI) techniques, aiming to ensure not only high levels of performance but also interpretability of the results. The models were trained and evaluated using publicly available brain MRI datasets, enabling the classification of different types of tumors. The results obtained demonstrated high Accuracy, Recall, and AUROC values, evidencing the effectiveness of the proposed solution in detecting brain tumors and confirming its potential as a clinical decision support system.
Descrição
Palavras-chave
Artificial Intelligence Medical Imaging Deep Learning Machine Learn- ing Image Segmentation Precision Medicine Inteligência artificial Imagens médicas Segmentação de imagens Medicina de precisão