TU Wien:Medizinische Bildverarbeitung VO (Langs)/Prüfung 2022-01-20

1.  a.) Which type of visual information of the training examples does an Active Shape Model (ASM) use?
    b.) Explain how a low-dimensional representation of the shape variability is created during the ASM training. 

2.  a.) Explain the difference between a General Linear Model (GLM) und Multi Voxel Pattern Analysis (MVPA) for the study of brain function and neuro imaging.
    b.) Which analysis method and which imaging modality do you have to use to detect brain areas that are active if a person moves a toe? What happens while the person is in the scanner?

3.  What is imaged with Magnetic Resonance Imaging (MRI)? Explain the basic principle of MRI.

4.  a.) Describe to properties of imaging data that is exploited by U-Nets to segment images.
    b.) Which input/output pairs of data are used for the training of  U-Net for the segmentation of cells in microscopy imaging data? 

5.  a.) Random Forests (RF) are classifiers, which use multiple variables to predict/classify a label. Explain how RFs can be used to score the relevance of individual features for the correct classification. 
    b.) What is the difference between a method to test the correlation of a single feature and the target variable (e.g., class) and the score you can derive from RF training?

6.  You are using an ROC curve to evaluate a classifier that detects malignant lesions in imaging data. The primary priority is to avoid false negatives in your detection. Which part of the ROC is relevant? Explain why.

7.  Explain how an algorithm for the detection of lesions in a liver with help of a classifier can be designed. Describe both training set, training, and the application of the algorithm to new data.

8.  Which (a) image similariate measure, and which (b) type of transformation is necessary for the image registration of PET and CT image volumes of the brain of the same patient? Please explain why. 

9.  Which imaging modality can be used to observe nerve fibre bundles?

10. Which method can you use to detect anomalies in images, even if during training you did not have access to any positive training examples (i.e., examples that have the anomaly)?