Saltar al contenido
República del Perú · Observatorio de la producción científica universitaria
OApol.org Observatorio de Inteligencia Científica Universitaria del Perú
OApol.org
Inicio Universidades Publicaciones Investigadores
Repositorios
Tesis y trabajos académicosRevistas
Indicadores
Ranking trimestralRanking por carreraObjetivos de Desarrollo SostenibleReporte OCDERed neuronal 3D
Acerca
MetodologíaPrivacidad

Different configurations for MobileNet in the detection of brain tumors

Gomez J.C.V.; Incalla A.P.Z.; Perca J.C.C.; Padilla D.I.M. · Proceedings of the 2021 IEEE 1st International Conference on Advanced Learning Technologies on Education and Research, ICALTER 2021 (2021) · DOI: 10.1109/icalter54105.2021.9675076

Resumen

Brain tumor is one of the most serious conditions associated with the brain, where a group of abnormal cells grows out of control. The field of computer vision has experienced a notable boom in the area of biomedical applications with the invention of different techniques in Deep Learning. The classification and detection of brain tumors is a primary task where convolutional neural networks are an alternative solution. But the main drawback of existing techniques is that they are complex with a large number of parameters that contribute to high execution time and high system specifications for implementation. In this article, it is proposed to use MobileNet with different settings in the classifier for the detection of brain tumors, a data set published in KAGGLE with a total of 4432 images has been used. The objective of the study is to obtain high efficiency with the MobileNet architecture that is of low complexity and to propose a system for the classification of brain tumors. With MobileNet as characteristics extractor, it was proposed to use four classifier configurations, as the first classifier, the Softmax function is used that obtained an accuracy of 96.63%, secondly, a Linear classifier was used that obtained an accuracy of 98.59%, in third place, SVM with a linear kernel was used as a classifier, which obtained an accuracy of 97.72%, and finally, SVM with a non-linear kernel was used as a classifier, which obtained 98.04%. With the results, it can be concluded that MobileNet is an excellent option to face this type of medical imaging problems due to its low computational cost and its high level of accuracy in detecting anomalies in the images provided.

Instituciones: Universidad Nacional de Moquegua