Recognize images that are not recognized correctly

Hello

I extract parts of the lungs from the CT scans, but in some cases this extraction is not done properly.

There are many images in the database. How do you think I can recognize and delete these images that have not been extracted correctly?

Thanks

[Dhruv G]

do you have ground truth data in some form?

[lech king]

yes

Answer 1

[Udit06]

There are multiple ways through which you can achieve the same:

1) Using basic statistics to detect anomalies such us unusually high or low pixel intensities in the images.

2) If the extraction process should result in a specific shape or size (e.g., lung shapes), you can use morphological analysis. You can refer to the following MathWorks documentation for more details related to morphological operations:

https://www.mathworks.com/help/releases/R2020b/images/morphological-filtering.html

3) Creating a machine learning or deep learning image classification for the same. You can refer to the following MathWorks documentation to train an image classifier:

https://www.mathworks.com/help/releases/R2020b/deeplearning/gs/create-simple-deep-learning-classification-network.html

I hope this helps.