Applied Operations Does not have ScaleIntensity

[a-parida12]

When applying the scale intensity operation(which does linear scaling), it would be nice to have some meta information stored regarding the original intensity values(like. the original min and the original max). This would be helpful in getting back the original intensity ranges after some reconstruction problems(like thru VAE, Super Resolution etc.)

[KumoLiu]

Hi @a-parida12, I'm not sure if I understand correctly. Since you already have the original image why need to get back to it? And for the Inverse transform, it allows, for example, images to be cropped, rotated, padded, etc., during training and inference, and after being returned to their original size before saving to file for comparison in an external viewer.
Thanks!

[a-parida12]

@KumoLiu thanks for the quick feedback. To clarify here are some additional information.

I am interested in the ScaleIntensity functionality. In a general workflow you load an MRI, which could be of variety of ranges, lets say [0, 1300] using ScaleIntensity I rescale it to [0, 1]. I pass it through a Super Resolution which produces an output in range [0, 1] too. When I send it to medical practitioner he expects the range to [0, 1300].

Currently, I re open the original image to get the range of the values. But it would be super nice to add the ScaleIntensity with an inverse too so helps prevent additional manual reading/tracking of original intensity ranges. Or at the least should have the original intensity values store in the meta information.

[KumoLiu]

@a-parida12, Oh I see. Maybe we can do something like sklearn.
@wyli @Nic-Ma, What do you think?

https://github.com/scikit-learn/scikit-learn/blob/f3f51f9b611bf873bd5836748647221480071a87/sklearn/preprocessing/_data.py#L513-L534

[a-parida12]

Just to clarify it would be something like the inverse method in https://docs.monai.io/en/stable/_modules/monai/transforms/spatial/array.html#Spacing ?

[AustinTapp]

Hi @a-parida12, @KumoLiu, @wyli, and @Nic-Ma. I was also wondering about an inverse for the ScaleIntensity transform. Could I ask what the status of this is, please?

Thank you!

[wyli]

There's another feature request here #5647 perhaps optionally track these applied parameters is a good starting point.