I had previously thought that the fractures will be given labels according to the sizes. But now to my knowledge...it's not true... How are they labelled then?

I had previously thought that the fractures will be given labels according to the sizes. But now to my knowledge...it's not true... How are they labelled then?  

  By: shirshak.acharya on July 28, 2024, 10:07 a.m.

Hello,

I have a small problem!

For example, if there are 6 broken bones in the left iliac then, the main largest bone would be label 11, then smaller than largest would be 12, accordingly , up to the smallest bone which will be label 16.

But now to my knowledge, it is not particularly the case, the label 14 can be the smallest fragment, and 
16 can be bigger fragment than 14 and 15 fragment. 

Now talking about prediction from model, The 2nd model (Fracture Segmentation Model) successfully predicts the main fragment, and remaining fragment.

(Label 0: bg, Label 1: main_fragment, Label 2: Remaining Fragment)

But now its the postprocessing part that i have a problem.

How to give labels for the broken separated fragments.

Can someone help?

Re: I had previously thought that the fractures will be given labels according to the sizes. But now to my knowledge...it's not true... How are they labelled then?  

  By: YSang on July 29, 2024, 1:39 a.m.

Hello! As this is an instance segmentation task, we will treat each detected fragments equally, and the order of the predicted fragments (within an ilium or sacrum) does not carry any meaning. It is okay to use any order as long as the anatomy assignment is correct. In the training data, the largest fragments happened to be assigned 1, 11, and 21.

In the evaluation, the predicted fragment instances and the ground-truth labels will be matched based on the highest IoU.

Best,

Yudi