%0 Conference Proceedings
%4 sid.inpe.br/sibgrapi/2021/08.31.13.14
%2 sid.inpe.br/sibgrapi/2021/08.31.13.14.58
%@doi 10.1109/SIBGRAPI54419.2021.00058
%T Enhancing Shallow Neural Networks Through Fourier-based Information Fusion for Stroke Classification
%D 2021
%A Roder, Mateus,
%A Rosa, Gustavo Henrique,
%A Papa, João Paulo,
%A Pedronette, Daniel Carlos Guimarães,
%E Paiva, Afonso ,
%E Menotti, David ,
%E Baranoski, Gladimir V. G. ,
%E Proença, Hugo Pedro ,
%E Junior, Antonio Lopes Apolinario ,
%E Papa, João Paulo ,
%E Pagliosa, Paulo ,
%E dos Santos, Thiago Oliveira ,
%E e Sá, Asla Medeiros ,
%E da Silveira, Thiago Lopes Trugillo ,
%E Brazil, Emilio Vital ,
%E Ponti, Moacir A. ,
%E Fernandes, Leandro A. F. ,
%E Avila, Sandra,
%B Conference on Graphics, Patterns and Images, 34 (SIBGRAPI)
%C Gramado, RS, Brazil (virtual)
%8 18-22 Oct. 2021
%I IEEE Computer Society
%J Los Alamitos
%S Proceedings
%K Stroke classification, Restricted Boltzmann Machines, Fourier transformation.
%X Deep learning techniques have been widely researched and applied to several problems, ranging from recommendation systems and service-based analysis to medical diagnosis. Nevertheless, even with outstanding results in some computer vision tasks, there is still much to explore as problems are becoming more complex, or applications are demanding new restrictions that hamper current techniques performance. Several works have been developed throughout the last decade to support automated medical diagnosis, yet detecting neural-based strokes, the so-called cerebrovascular accident (CVA). However, such approaches have room for improvement, such as the employment of information fusion techniques in deep learning architectures. Such an approach might benefit CVA detection as most state-of-the-art models use computer-based tomography and magnetic resonance imaging samples. Therefore, the present work aims at enhancing stroke detection through information fusion, mainly composed of original and Fourier-based samples, applied to shallow architectures (Restricted Boltzmann machines). The whole picture employs multimodal inputs, allowing data from different domains (images and Fourier transforms) to be learned together, improving the model's predictive capacity. As the main result, the proposed approach overpassed the baselines, achieving the remarkable accuracy of 99.72%.
%@language en
%3 Paper ID 12.pdf