hi！recently i have been working on no-contrast CTs(more than 300) kidney segmentation，while 90% of them get very bad performance on KiTS.
it confused me and I have checked the two datasets ，both KiTS and mine ，to ensure they remain highly consistent expect for differences in kidney HU intensity.
I do have some other CTs with contrast (just like the KiTS) and the fact is that when do the inference on kidney filled with medium, KiTS model demonstrats outstanding performance but it fails to segment a whole kidney in no-contrast CTs.
is there any chance that KiTS21/19 model rely on high HU intensity too much to make judgments using contrast，while hardly capturing other features such as position and shape so that when medium disappears the model cannot recognize kidneys?