TensorFlow 2.0 implementation of a model predicting region of interest in MRI localizers.
Link to paper: https://www.mdpi.com/2306-5354/10/1/92
The model can be trained to prediction the left-right boundries or top-bottom boundries of an ROI for a stack of hip or abdomen MRI localizer. Two instances of the same model is trained independently for the left-right and top-bottom pairs of boundries. The input to the model is a stack of 512x512 2D 1-channel grey images, and the output of the model is two scalars indicating the coordinates within the 512 pixels.