Currently, I am trying to use MONAILabel to train a model for air segmentation of MR images in the pelvic region. The approach I used was : 1) Create an already air annotated dataset
2) Use the radiology app and train a model from scratch.
However, the model accuracy using the default settings (e.g. random active strategy) is limited in 0.7 % which is low in my case.
Do you know any more detailed way of increasing the model performance? One way I thought would be to create more annotated data and use it a training dataset. However, this is time consuming.
What do you suggest?
Thank you for your time in this.