Análisis del comportamiento de arquitecturas de redes neuronales convolucionales en la detección de malaria

Abstract

Health and technology are closely related today, because thanks to technological advances, various solutions have been achieved to the different problems that are immersed in health, one of these problems is the detection of malaria in a timely manner for their respective treatment. In this work, a deep learning is carried out with the Mobilenet and Inception V3 models, using a Kaggle database called Malaria Cell Images Dataset, which is constituted by two Parasitized classes that are the images that show signs of existence of The infection of Malaria and Uninfected that does not present anomalies, each class contains 13,780 images that give a total of 27,560 images, the technological tools used were Python as a programming language under its Anaconda distribution and Jupyter environment and the Keras and TensorFlow libraries. With this, an analysis of the behavior of both architectures must be performed, for which approximately 8,000 images are considered, because a large number of images requires very powerful computational equipment and because of the quality of images contained in the data set, since it does not have a good resolution, therefore, the accuracy obtained in both models was 0.9894 and 0.9555, respectively.