Please use this identifier to cite or link to this item:
Title: Mejora para la imagen de Rayos X mediante el uso de Deep Learning
Author: Moreno Berdón, Patricia
Director: Alférez Baquero, Edwin Santiago
Tutor: Calvet Liñan, Laura  
Keywords: X-rays
image enhancement
deep learning
Issue Date: Jun-2021
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: X-ray imaging is an important diagnostic tool. However, despite its clinical value, it offers poor soft tissue contrast as well as a high radiation dose. That is why improving the contrast and trying to reach a minimum dose with an image quality compatible with the diagnosis is necessary. Conventional enhancement methods for radiological imaging are based on the concatenation of solutions for each of these two problems (contrast and noise), providing good results but with a high computational cost that makes real-time processing, necessary in clinical radiography, difficult. In this work, we proposed a method that integrates contrast enhancement and noise reduction for radiological images through deep learning, eliminating the need to find the appropriate parameters for the different studies and having the processed image in less than a second. 6 A bibliographic study of new methods based on deep learning for noise and contrast improvement was made, selecting the UNet architecture with different cost functions and different convolutional neural network encoder architectures. After visual evaluation of the results, it is established that the best cost functions are the multiscale structural similarity index and the multiscale structural similarity index combined with the mean absolute error. Among the tested architectures for the encoder, those that provide the best results are ResNet34 and EfficientNetB3. Therefore, a new method based on deep learning has been proposed that allows to improve the contrast and reduce the noise of animal radiology images in less than 1s.
Language: Spanish
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
pmorenobeTFM0621memoria.pdfMemoria del TFM2.78 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons