Reconocimiento de la presencia de sars-cov-2 en pulmones a través de imágenes de radiodiagnóstico haciendo uso de machine learnig con lenguaje de programación Python

Abstract

The goal of this project was to use Machine Learning (ML), for SARSCoV-2 recognition, using medical images acquired by computed tomography of the chest region in DICOM format, from a 2-slice Siemens somatom tomograph and a data set in the cloud, which were subsequently transformed to "png" images. The recognition system was built using the "Python" programming language, making use of open-source libraries, both for Machine Learning and "TensorFlow"; "Pydicom" was used to manage DICOM files and "Open CV" for images. The images were imported into a pre-trained convolutional neural network, adapting it to the type of multi-class classification of the project, applying data augmentation techniques (Data Augmentation), exponential decays of neural network parameters such as the Learning Rate, training the convolutional neural network, optimising the appropriate parameters for its correct recognition operation, subsequently a web interface was developed using the "Streamlit" library for the management and applicability of the model being of dynamic use for the user and being multiplatform. Quantitative results were obtained that reflected the effectiveness of the model with an efficiency of 88% for detecting COVID-19. The prior installation of Python libraries is recommended for the correct functioning of the recognition system.