prediction about 2d-denseUnet

[GY2019]

hello~ how could we use the trained 2d-denseUnet model to predict the liver areas. Should we change the predict_tumor_inwindow function in test.py? thanks~

[hongson23]

Hello @xmengli999, I have the same question. Could we predict liver segmentation using 2d-denseUnet? Thank you

[xmengli]

The liver area is already generated and you can download "livermask". The 2d-denseUnet model used to produce accurate liver and tumor mask.

@hongson23 yes, you can.

[GY2019]

thanks for your reply. because we want to have a Transfer learning based on your dataset. so we want to know how can we use 2d-model to produce liver mask or predict the liver areas. thanks a lot

[xmengli]

To produce the liver segmentation, I suggest you to use a standard 2D ResNet/denseunet model and you can get a satisfactory results(at least higher than 90%). You can remove the tumor label, and change the channel of the last convolution layer to 2.

[GY2019]

ok~ i get your suggestion , change the class from 3 to 2 . but how to write the prediction function? load the trained model and load the "predict_window_mulgpu" in the ./lib/func.py ?