How to turn on active learning - epistemic uncertainty

[hw-ju]

Hi there,

I'm using the latest stable versions of monai label (both v0.4.2 installed by pip and v0.4.0 by docker image), and try to turn on epistemic uncertainty in DeepEdit for radiology. I modified the config file of DeepEdit, change to true from false:

After that, when I start the server, the output shows it is turned on

[2022-07-25 16:23:42,642] [20251] [MainThread] [INFO] (lib.configs.deepedit:144) - EPISTEMIC Enabled: 1; Samples: 5

but no calculation of the epistemic score, and in 3DSlicer, there's no EPISTEMIC option in the Active Learning section.

After I click Next Sample in 3DSlicer, I see this in server output:

[2022-07-25 23:45:07,511] [381] [MainThread] [INFO] (monailabel.endpoints.activelearning:42) - Active Learning Request: {'strategy': 'random', 'client_id': 'user-xyz'}

How to turn on active learning?

[diazandr3s]

Hi @hw-ju,

This is a good question.
Please add this line when starting the server:

--conf skip_scoring false --conf skip_strategies false --conf epistemic_enabled true

Otherwise, change those flags within the main file: https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/radiology/main.py#L153

[hw-ju]

@diazandr3s That works! Thanks!
Just to confirm, after using that line, we also need to choose epistemic in MONAI Label 3DSlicer plugin as in screenshot below, correct? I tried choosing random option in the plugin, and it began fetching randomly.

[diazandr3s]

That's correct! Thanks for the update :)