Fix callouts and adapt texts

README.md

@@ -7,21 +7,26 @@
 [![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 <!-- [![codecov](https://codecov.io/gh/BrainLesion/deep_quality_estimation/graph/badge.svg?token=A7FWUKO9Y4)](https://codecov.io/gh/BrainLesion/deep_quality_estimation) -->
 
-Quality prediction for brain tumor segmentation on a scale ranging from &#x2B50; 1 star to  &#x2B50;&#x2B50;&#x2B50;&#x2B50;&#x2B50;&#x2B50; 6 stars inspired by the paper [**Deep Quality Estimation: Creating Surrogate Models for Human Quality Ratings**](https://arxiv.org/abs/2205.10355).  
+Quality prediction for brain tumor segmentation on a scale ranging from &#x2B50; 1 star to  &#x2B50;&#x2B50;&#x2B50;&#x2B50;&#x2B50;&#x2B50; 6 stars inspired by the paper [**Deep Quality Estimation: Creating Surrogate Models for Human Quality Ratings**](https://arxiv.org/abs/2205.10355).  <br>
 This can be used to estimate the quality of a BraTS glioma segmentation for evaluation purposes or, e.g., as part of a loss function during model training.
 
-> [!NOTE]  
+
+## Important notes
+
+> [!IMPORTANT]  
 > This package expects images in atlas space and segmentation labels in brats style, i.e.
 > - `label 1` is the necrotic and non-enhancing tumor core
 > - `label 2` is the peritumoral edema
 > - `label 3` is the GD-enhancing tumor (used to be `label 4` in older data; both are supported)
 
-
 > [!NOTE]
-The model in this package differs from the one presented in the paper. Instead, it is trained based on individual radiologists' ratings. This way, the model can learn the variance between radiologists' estimates. It outperforms the model presented in the paper on the test set.
+> The model in this package differs from the one presented in the paper. <br>
+> Unlike the original model it is trained based on individual radiologists' ratings enabling it to learn the variance between radiologists' estimates. <br>
+> It outperforms the model presented in the paper on the test set.
 
-> [!NOTE]
-The model is biased to overestimate segmentation quality as it was mainly trained on high-quality segmentations and was exposed to only a few bad samples.
+> [!CAUTION]
+> The model is biased to overestimate segmentation quality as it was mainly trained on high-quality segmentations and was exposed to only a few bad samples.
+> We still argue that high scores can be useful.
 
 
 ## Installation