What is the correct way to calculate the average dice score for multiple data in binary image segmentation?

How to calculate the average dice score in binary image segmentation for multiple data? Should I calculate the dice score for each image separately and then average them? or do I need to calculate a single dice score over the total "TP", "FP", and "FN" values for all images? Which one is the correct way?

Answer 1

[KALYAN ACHARJYA]

Here>explanation with references

https://stats.stackexchange.com/questions/550646/proper-way-to-calculate-mean-dice-coefficient-on-a-dataset

[Ahmet Ilhan]

Thank you for the answer Mr. Acharjya. I found an answer about this that contradicts the reference you shared from another form page. What is your opinion about that?

https://stats.stackexchange.com/questions/519508/correct-way-of-computing-dice-score-for-image-segmentation