py-SAINT 1.3.0

SAINT

This is SAINT. Environment configuration: pytorch:0.4.1 python:3.6 cuda:9.0 SAINT: Spatially Aware Interpolation NeTwork for Medical Slice Synthesis step1: .nii->.pt from py_SAINT.STAGE1 import nii2pickle nii2pickle.nii2pt(ori_dir_path,output_file_path) eg: nii2pickle.nii2pt("/home1/mksun/xh_data/273data-yscl/1T2/1/002_OCor_T2_FRFSE/","/home1/mksun/SAINT/Data/Stage1_Input/TEST/HR/")

       ori_dir_path:  file path containing .nii
       output_file_path:  file path of the generated .pt

step2: Interpolate with sag and cor view respectively from py_SAINT.STAGE1 import interpolation interpolation.get_Stage1_result (scale ='4',save =/path/ ,dir_data ='/path/',n_colors =3 ,n_GPUs =1,rgb_range =4000, view ='sag',gpu='0') interpolation.get_Stage1_result (scale ='4',save =/path/ ,dir_data ='/path/',n_colors =3 ,n_GPUs =1,rgb_range =4000, view ='cor',gpu='0') eg:interpolation.get_Stage1_result (scale ='4',save ="/home1/mksun/SAINT/Data/Stage1_output_sag_cor/" ,dir_data ='/home1/mksun/SAINT/Data/Stage1_Input/',n_colors =3 ,n_GPUs =1,rgb_range =4000, view ='cor',gpu='0') eg:interpolation.get_Stage1_result (scale ='4',save ="/home1/mksun/SAINT/Data/Stage1_output_sag_cor/" ,dir_data ='/home1/mksun/SAINT/Data/Stage1_Input/',n_colors =3 ,n_GPUs =1,rgb_range =4000, view ='sag',gpu='0')

       scale:  super resolution scale (eg:2,3,4,6)
       save:  file path of save           
       dir_data:  dataset directory (Note that the dir_data path should point to a folder that contains subfolders named  'TEST', each of which needs to have a 'HR' and 'LR' subfolder, 
                   'HR' is high resolution file , 'LR' is low resolution file. Data should go accordingly in this structure.
       n_colors:  number of channels to use
       n_GPUs:  number of GPUs
       rgb_range:  maximum value of RGB
       view:  view of interpolation (Note the --view option performs inference on the volume from either the sagittal or coronal axis. Note that the whether it's actually sagittal or coronal depends on the oritentation of the data.)

step3: Before going to the RFN stage, sagittal and coronal-wise SR'ed volume needs to be recombine into a single volume for inference. In simple terms just concatenate them in the first dimension, coronal SR goes in channel 0 and sagittal SR goes in channel 1. from py_SAINT.STAGE1.process import cor_sag_comb_test cor_sag_comb_test.comb_cor_sag(files_dir='/path/',input_sag_cor_dir='/path/',out_dir='/path/', scale=4) eg:cor_sag_comb_test.comb_cor_sag(files_dir='/home1/mksun/SAINT/Data/Stage1_Input/TEST/HR/',input_sag_cor_dir='/home1/mksun/SAINT/Data/Stage1_output_sag_cor/results/raw/',out_dir='/home1/mksun/SAINT/Data/combine_cor_sag_out/TEST/', scale=4) files_dir: dataset directory input_sag_cor_dir: path to the folder containing sag and cor out_dir: generated combine path scale: super resolution scale step4: Residual-Fusion from py_SAINT.STAGE2 import fuse fuse.get_Stage2_result(save ='/path/',dir_data ='/path/' ,n_GPUs =1 ,rgb_range =4000,gpu='0') eg: fuse.get_Stage2_result(save ='/home1/mksun/SAINT/Data/out_fuse/',dir_data ='/home1/mksun/SAINT/Data/combine_cor_sag_out/' ,n_GPUs =1 ,rgb_range =4000,gpu='0') save: file path of save
dir_data: step3_out_dir n_GPUs: number of GPUs rgb_range: maximum value of RGB (option): .pt->.nii from py_SAINT.STAGE1 import pt2nii pt2nii.pt2nii(ori_nii_dir_path, pt_dir_path,nii_dir_path) nii_dir_path: nii_output_dir eg锛歱t2nii.pt2nii(ori_nii_dir_path='/home1/mksun/xh_data/273data-yscl/1T2/1/002_OCor_T2_FRFSE/',pt_dir_path='/home1/mksun/SAINT/Data/out_fuse/results/raw/',nii_dir_path='/home1/mksun/SAINT/Data/final_nii/')