Is it hard to detect an object without boundary edge?

Hi I am trying to train YOLO v1 model, which detects some kind of crack in human pelvic bone.

That crack thing appears on the same spot in pelvic bone.

I have about 50000 images and bounding box pair and trained it for 100 epochs

but the model performance is little disappointing.

So, I am assuming that if the object (crack in this case) doesn’t have exact boundary edge in the image,

the model won’t be train as I thought.

Is it right?