Detecção e Classificação de Câncer a partir de Mamografias Digitalizadas e Redes Neurais Convolucionais

• Autor(a) principal: Dalvi, Rodolfo de Figueiredo
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Espírito Santo (riUfes)
• Texto Completo: http://repositorio.ufes.br/handle/10/10177
• topic: Breast cancer; Mammography; Computer Aided Diagnosis; Segmentation; Convolutional neural network; Rede neural convolucional; Câncer de mama; Diagnóstico assistido por computador; Segmentação; Eletrônica Industrial, Sistemas e Controles Eletrônicos; Mamas - Câncer - Diagnóstico por imagem; Mamografia; Processamento de imagens auxiliado por computador; 621.3

In this work, two methodologies based on convolutional neural networks were developedwith the objective of performing the classification of breast cancer in a digital mammographyimage. Firstly, we sought a simpler methodology that, from regions marked in the database,identifies whether they are normal or carcinogenic regions. In this way, expressive resultswere obtained, with an accuracy of 99.41%, sensitivity and specificity of 98.57% and100%, respectively. In order to seek greater applicability of the proposed methodology,an improvement of the first methodology was made, making it independent of an initialmarking. In the second methodology is performed a pre-processing of the image andthen segment the potential candidates for cancer. Each candidate is classified as normalor carcinogenic by a convolutional neural network. Finally, carcinogenic candidates areclassified as benign or malignant, also using a convolutional neural network. Relevantresults were obtained considering the difficulty of the problem, reaching an accuracyof 91.89%, sensitivity of 88.52% and specificity of 96.00% for cancer detection, and anaccuracy of 82.14% , sensitivity of 81.48% and specificity of 82.75% for the classificationof the type of cancer.