Skip to content

Changes

Page history
Update Upscaling authored by Bartosz Sobierajski's avatar Bartosz Sobierajski
Show whitespace changes
Inline Side-by-side
Upscaling.md
View page @ ded8a857
...@@ -2,6 +2,7 @@ ...@@ -2,6 +2,7 @@
title: Upscaling title: Upscaling
--- ---
This page provides an overview of the upscaling component in the KI-Vol 2024 project, covering the motivation, relevant concepts in MRI image acquisition, challenges, and techniques employed. The primary objective of this component is to enhance MRI image resolution through deep learning, specifically focusing on super-resolution networks. By improving resolution, this approach enables the use of lower base resolutions during scanning, effectively reducing MRI acquisition time, which is crucial for patient comfort and accessibility. In our research, we trained and tested two prominent models, SRResNet and SRGAN, for upscaling MRI images, assessing their effectiveness in enhancing the quality of low-resolution images derived from downsampled MRI data. This page provides an overview of the upscaling component in the KI-Vol 2024 project, covering the motivation, relevant concepts in MRI image acquisition, challenges, and techniques employed. The primary objective of this component is to enhance MRI image resolution through deep learning, specifically focusing on super-resolution networks. By improving resolution, this approach enables the use of lower base resolutions during scanning, effectively reducing MRI acquisition time, which is crucial for patient comfort and accessibility. In our research, we trained and tested two prominent models, SRResNet and SRGAN, for upscaling MRI images, assessing their effectiveness in enhancing the quality of low-resolution images derived from downsampled MRI data.
# Motivation # Motivation
...@@ -148,7 +149,7 @@ To assess the effectiveness of the chosen upscaling method, we focus on the eval ...@@ -148,7 +149,7 @@ To assess the effectiveness of the chosen upscaling method, we focus on the eval
![Base Fold 1 Evaluation](https://gitlab.uni-koblenz.de/ki-vol/ki-vol/-/raw/main/Images/base_evaluation.png?ref_type=heads) ![Base Fold 1 Evaluation](https://gitlab.uni-koblenz.de/ki-vol/ki-vol/-/raw/main/Images/base_evaluation.png?ref_type=heads)
<figcaption> <figcaption>
<table style="width: 100%; text-align: center;"> <table>
<tr> <tr>
<th>Dice TC</th> <th>Dice TC</th>
<th>Dice WT</th> <th>Dice WT</th>
...@@ -170,6 +171,8 @@ To assess the effectiveness of the chosen upscaling method, we focus on the eval ...@@ -170,6 +171,8 @@ To assess the effectiveness of the chosen upscaling method, we focus on the eval
</figure> </figure>
</div> </div>
The original valitation data from Fold 1 (as seen in Figure 8) will be used for comparison.
```python ```python
# Data transforms validation # Data transforms validation
factor = 4 factor = 4
......