Feature extraction#

In this notebook, you will create an instance segmentation of biological data and extract quantitiative features from this data with the regionprops_table() function from scikit-image.

from skimage import data, filters, measure
import pandas as pd
import matplotlib.pyplot as plt

Different types of features#

As shown in the lecture, features can be grouped in a few general types. Many features tyically belong to one of the following:

  • Intensity-based features: These are based on the image intensity values in selected areas of interest

  • Shape-based features: These describe the general shape of an object and can be measured independ of the original intensity values

  • Spatial features: These features typically take into account not only the object itself but also its location in the image or with reference to other objects.

Working with dictionaries#

Measured features of an image are essentially tabular data, which can be handled very efficiently in Python. Tabular data for typical labelled image data looks like this:

Label

feature 1

feature 2

1

some value

some value

2

some value

some value

3

some value

some value

Remember: Labelled images with multiple occurrences of the same type of objects (e.g., cells or nuclei) are the result of an instance segmentation task, whereas a unique label is assigned to every object.

Recap: Dictionaries#

Dictionaries in Python are a handy datatype to keep track of mixed data (strings, numbers, etc). They have a key-value structure and can be created and accessed like this:

data1 = [1, 2, 3]
data2 = ['Monday', 'Tuesday', 'Wednesday']
my_dict = {'numbers': data1,
          'days': data2}
my_dict
{'numbers': [1, 2, 3], 'days': ['Monday', 'Tuesday', 'Wednesday']}
my_dict['days']
['Monday', 'Tuesday', 'Wednesday']
my_dict.keys()
dict_keys(['numbers', 'days'])

The Pandas library provides a great amount of useful functions to work with tabular data. The pandas-equivalent of a dictionary is called a DataFrame and can be created from a dictionary by simple means:

df = pd.DataFrame(my_dict)
df
numbers days
0 1 Monday
1 2 Tuesday
2 3 Wednesday

Accessing the data works just like with dictionaries:

df['days']
0       Monday
1      Tuesday
2    Wednesday
Name: days, dtype: object

Exercises#

First, let’s get some sample data from scikit-image:

image = data.human_mitosis()
plt.imshow(image, cmap='gray')
<matplotlib.image.AxesImage at 0x202d039f640>
../_images/01_Feature_extraction_11_1.png

Exercise 1#

Apply Otsu-thresholding to the image to create a binary image and then create a label image from the binary image:

Exercise 2#

Use the measure.regionprops_table function to measure the area and the mean intensity for every object.

Hint: You can create a list of measurements to be passed to regionprops_table like this: properties = ['property1', 'property2', ...]. You find a list of all possible properties here.

results = 

Exercise 3#

The results variable now contains the derived measurements from the input data and is of type dict.

  • Use the dictionary.keys() command to print all columns in the results variable to the notebook.

  • Remember, Python dictionaries can be accessed like this: value = dictionary[key]. Use this this to print all area measurements from results to this notebook!

Exercise 4#

When we obtain measurements from image data, we usually want to visualize them and do some statistical evaluation. You have already learned to plot histograms with matplotlib - use this to visualize the distribution of areas in the image data as a histogram!

Hint: You can retrieve the area measurements from the results as described above.

Exercise 5#

Lastly, calculate a mean and standard deviation of the areas of all objects in the image. In order to do so, you need to

  • retrieve the measurements from the results dataframe as described above

  • Convert it to a numpy array with np.asarray()

  • Calculate and print the mean and the standard deviation from the results

Hint: Numpy arrays have some convenience functions attached to them (e.g., some_array.function()) attached to them - see if you can find out the functions for the mean and standard deviation!