Diagnóstico de covid-19 mediante el análisis de radiografías pulmonares empleando un modelo basado en redes neuronales convolucionales (CNN)
Since the declaration of the health emergency caused by Covid-19 in March 2020, approximately 219 million people have been infected to date, of which 4.5 million have died. In Ecuador, it is estimated that there are 508,000 confirmed cases and approximately 32,000 deaths due to this disease. Despite...
Gorde:
| Egile nagusia: | |
|---|---|
| Formatua: | bachelorThesis |
| Hizkuntza: | spa |
| Argitaratua: |
2022
|
| Gaiak: | |
| Sarrera elektronikoa: | https://dspace.unl.edu.ec/jspui/handle/123456789/24495 |
| Etiketak: |
Etiketa erantsi
Etiketarik gabe, Izan zaitez lehena erregistro honi etiketa jartzen!
|
Etiketa erantsi | Gaia: | Since the declaration of the health emergency caused by Covid-19 in March 2020, approximately 219 million people have been infected to date, of which 4.5 million have died. In Ecuador, it is estimated that there are 508,000 confirmed cases and approximately 32,000 deaths due to this disease. Despite the availability of verified methods to diagnose Covid-19, PCR or RT-PCR tests tend to generate false positives and negatives between 30% and 40%. Therefore, helping traditional methods to make an accurate clinical diagnosis, using lung radiographs as input data, represents a radical change in the detection of Covid-19, since it is a much more comfortable alternative for the patient and, more importantly, increases the level of accuracy while reducing false positive and false negative rates. For this reason, in this degree work, we propose the creation of a model based on the Convolutional Neural Network (CNN) architecture, capable of analyzing pulmonary radiographs for the diagnosis of Covid-19. In order to obtain the raw material for the model, a search protocol based on related works was carried out, obtaining as a result a complete and varied data set with lung radiographs suitable for use by the model in the training and test phase. Regarding the construction phase of the diagnos19 model, VGG-16 was chosen as the base model, to which a fine tuning of its hyperparameters was performed using the fine tuning technique, which, after validating and evaluating the results by applying techniques such as zero shot learning and human validation, resulted in an accuracy level of 99. 167%, with a sensitivity level equal to 99.167%, with these results it was possible to exceed the 90% accuracy established in the research question, in addition, our results are comparable with the values obtained in the models of the related works, whose highest value in terms of accuracy was 98.27%, with a sensitivity level equal to 98.93%. Key words: Covid-19, CNN, VGG16, lung radiographs, X-rays. |
|---|