Mejora del diagnóstico médico de radiografías de tórax usando aprendizaje profundo con aumento gradual de iteraciones

Abstract

Pneumonia is an inflammatory condition of the lung that affects the alveoli. Diagnosis is based on symptoms and physical examination. Chest radiographs are used as an alternative to validate the diagnosis. In the present work, a methodology is presented to perform image processing based on machine learning and artificial intelligence to perform an automatic classification of said images. Results of experiments carried out in two classification scenarios are presented: crossvalidation and training and test sets. Five different machine learning methods were used in each classification scenario, as well as five evaluation metrics. Similarly, the images were preprocessed with five filters, in addition to the original images. The oriented gradient histogram feature descriptor was used to measure the effectiveness in both cases: original and with filters. The configuration of the experiment was planned in such a way that it allowed to identify the best classification conditions, also allowing to clearly observe the impact of the size of the training set on the evaluation metrics used. The results obtained allow us to see the effectiveness of the implemented methodology, since the results are competitive with those reported in the state of the art.