Application of Super Resolution Convolutional Neural Networks (SRCNNs) to enhance medical images resolution
| Autor: | Lorenzo Polo, Andrea |
|---|---|
| Přispěvatelé: | Pascau González-Garzón, Javier, Calvo Haro, José Antonio, Universidad Carlos III de Madrid. Departamento de Bioingeniería e Ingeniería Aeroespacial |
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: | |
| Zdroj: | e-Archivo. Repositorio Institucional de la Universidad Carlos III de Madrid instname |
| Popis: | The importance of resolution is crucial when working with medical images. The possibility to visualize details lead to a more accurate diagnosis and makes segmentation easier. However, obtention of high-resolution medical images requires of long acquisition times. In clinical environments, lack of time leads to the acquisition of low-resolution images. Super Resolution (SR) consist in post-processing images in order to enhance its resolution. During the last years, a branch of SR is getting promising results. This branch focuses in the application of Convolutional Neural Networks (CNNs) to the images. This project is intended to create a network able to enhance resolution of knee MR stored in DICOM format. Different networks are proposed, and evaluation is made by computing Peak Signal-to-Noise Ratio (PSNR) and normalized Cross-Correlation. One of the networks proposed, SR-DCNN, presented better results than the conventional method, bicubic interpolation. Finally, visual comparison of the SR-DCNN and bicubic interpolation also showed that the network proposed outperforms the conventional methods. Ingeniería Biomédica |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|