Practice Exam Questions

The signal to noise ratios for the image are very high (>3) in 1 and 4, reasonable in 2, and completely unusable in 4. The images in 1,2 and 4 can be processed using a Gaussian filter and threshold while might have a tiny bit of information but would need to be heavily filtered and interpreted carefully. It is easiest therefore, to examine the histograms of the images first where 3 has a much broader peak (higher standard deviation) indicating it is noisier and skipping this image in the first place.

A threshold cannot be applied to the image because there is significant structure in the background. The first step is to correct this structure and then apply a threshold. If this is not possible edge enhancement filters can be applied to improve the image and make the different phases easier to distinguish.

For shape anisotropy means that an object is not dimensioned the same in all directions: in some directions it is larger or stronger than others. Conversely an isotropic shape is the same in all directions (a circle or sphere). For cells responsible for mechanosensation in bone, an anisotropic shape is an indication of which direction the cell has the most sensitivity.

To validate the image, you could do a number of different tests. The easist would be to manually identify the cells in the image and compare the manual identification to the segmentation. The best approach would be to analyze the same cells with another technique (fluorence imaging, protein screening) and compare the relevant metrics. In some cases this is cell positions, but a volume fraction or other ensemble values are also sufficient. To further improve the segmented images, a number of the approachs discussed in Advanced Segmentation could be used ranging from basic morphology (an open would remove single voxel noise) to component labeling with a minimum voxel count (anything small than 50 voxels is not a cell) to shape based criterian (only objects with anisotropy between 0.5 and 1 are cells)

There is no evidently emergent pattern in the cell distribution and they appear to be more or less randomly placed. The randomness of the placement could be characterized by examining the self-avoiding factor (or clustering factor) of the system based on the nearest neighbor distance and the density of cells.

While the positions appear to be random, there is a pattern to the placement such that some regions are denser than others. The distribution of the placement could be characterized by examining the self-avoiding factor (or clustering factor) of the system based on the nearest neighbor distance and the density of cells. Furthermore the 2D (or 3D) characteristic could be investigated by calcuating the ensemble distribution tensor in different regions and comparing them. The ellipsoidal representation of this tensor will be smaller in the denser regions and larger in the sparser regions.