Hi yasmin ismail,
I understand you want to choose a threshold based on dice score for a binary image segmentation problem. According to my understanding, the dice score is a metric applied between two binary images. It provides a score that increases with increase in similarity between two images.
To choose dice threshold, please perform the following steps:
1. Firstly, you will need the segmentation output for each pixel of an input image. This is a score between [0,1] reflecting the confidence of the prediction.
2. To perform segmentation of the image, consider a range of threshold values between [0,1]. Use a for loop iterate through each threshold value and binarize the image.
3. Binarization can be done by marking the pixels with scores greater than the threshold as 1 and less than the threshold as 0. Binarization can be performed by using the MATLAB function “imbinarize(I, T)” where I is the original image and T is the binarization threshold.
For more information on “imbinarize” function, refer to the link of MathWorks documentation given below:
https://www.mathworks.com/help/images/ref/imbinarize.html?s_tid=doc_ta
4. For each binarized image corresponding to one threshold value compute the dice score between the binarized image and the ground-truth binarized image. The dice score can be computed using the MATLAB inbuilt function “dice”.
dice(BI1, BI2) % BI1 and BI2 are binary images
Refer to MathWorks documentation for more information on “dice” function:
https://www.mathworks.com/help/images/ref/dice.html?s_tid=doc_ta
5. Next, plot the dice-score as a function of threshold values. Finally, choose the threshold value that maximizes the dice-score.
Refer to MathWorks documentation for more information on “plot” function:
Hope this helps!
