Overview

0. Introduction
1. Thresholding
2. Segmentation
3. Otsu's Method
4. Edge Detection
5. Canny Edge
6. Outlook and Conclusion

2: Segmentation

For your second task, you will be implementing a plugin, which will be capable of evaluating the quality of a segmentation by comparing it to a given reference-image. In clinical practice these reference images can be attained through manual segmentation of the image by a skilled physician.

In your case the provided reference image will look like this:

---

2.1: A new class

In order to quantify the quality of your segmentation, you will be calculating the values for Sensitivity and Specificity, which were introduced in your blackboard-exercise. In addition, you will be creating a dedicated class to store and access these values.

To do:

1. Create a new class in the src-Folder named EvaluationResult
2. Create the class-variables sensitivity and specificity (both double)
3. Implement a constructor:

   public EvaluationResult ( double specificity , double sensitivity ){}
   

4. Implement getter-methods for both values:

    public double getSpecificity (){}
    public double getSensitivity (){}
   

---

2.2: The Plugin

The rest of the code for this task will be implemented in the provided class Task_2_EvaluateSegmentation.

In order to calculate the values for Sensitivity and Specificity, you will need to assign a "state" to each pixel of your segmented image. The states in question are:

State
TP	"True Positive" → the pixel was correctly identified as part of the target
TN	"True Negative" → the pixel was correctly identified as background
FP	"False Positive" → the pixel was falsely identified as part of the target
FN	"False Negative" → the pixel was falsely identified as background

For decerning which of these cases applies to a given pixel, the reference image is used as the "correct" segmentation.

The values for Sensitivity and Specificity can then be calculated as follows:

To do:

1. Create a new method:

    private EvaluationResult evaluateSegmentation ( ImageProcessor
    segmentation , ImageProcessor reference ){}
   

2. Check if both images have the same dimensions - if not return null
3. Iterate over both images and count up the number of occurrances for each state
4. Calculate the values for Sensitivity and Specificity using the formulas listed above
5. Create a new EvaluationResult-object to store these values and return it

---

Now that you have created a method which perfroms the actual evaluation, you will need to implement the run-method in order to apply the plugin to an image.

To do:

1. Use the IJ.openImage()-method to open a window, which allows the user to select the reference image.

2. Check wether the reference image has been loaded successfully - if not throw a fitting exception

   💡 Tip:
   

   Check the ImageJ-API to see how the method behaves when no image has been loaded

3. Once the image was loaded successfully, apply your evaluateSegmentation()-method and print your results

   📝 Note:
   

   As a test for your code, you could for example choose the cells_reference image as both "segmentation" and "reference". The plugin should then return 1.0 for both values.

---

2.3: Project-Report

The part of your report concerning Task 3 should contain the following:

• A brief explanation as to what TP, TN, FP, FN mean
• The formulas for Specificity and Sensitivity
• A brief explanation on what Specificity and Sensitivity mean in the context of image-segmentation

Additionally you should also include your own results from this exercise:

• Using your plugin from Task 1 generate 6 different segmentations of the "cells"-image using approapriate hyperparamters.
• Use the plugin you implemented in this task to evaluate each segmentation
• Display your results in an appropriate way
• Name two other ways to evaluate how good the segmentation is

Next