Can I generate my own brand new segmentation model by training without loading the pre trained model?

I would like to use MonAILabel to create a new model for segmenting the bone gray matter of vertebrae L1, L2, and L3. May I ask if I have modified the self labels in{
“Background”: 0,
“SegmentionL1”: 1,
“SegmentionL2”: 2,
“SegmentionL3”: 3,
}And if strtobool
(self. conf. get (“use_pretrained_model”, “false”):,
How to train without loading pre trained models and generate your own brand new segmentation model? In fact, after trying it out, I found that it did not work