Multilable segmentation with DiceMetric

[greenyjwk]

Hello,

I want to find each class dice score using DiceMetric instead of finding whole Dice score.

My prediction dimension is (1,4,255,255,255 ) == (Batch, Class, Width, Height, Depth)
and
My label dimension is (1,1,255,255,255 ) == (Batch, , Width, Height, Depth)
Background:0, Kidney:1, Tumor:2, Cyst:3

Using DiceMetric(monai.metrics.DiceMetric) I want to find the dice score for only Tumor dice score. Which arguments would I need to change from DiceScore for this task?

[KumoLiu]

Hi @greenyjwk, you can set reduction="mean_channel" to get dice for each channel.

MONAI/monai/metrics/meandice.py

Line 42 in 4029c42

reduction: define mode of reduction to the metrics, will only apply reduction on `not-nan` values,

Hope it helps, thanks.