search

catactg / lasc

Left Atrial Segmentation Challenge 2013
BSD 2-Clause "Simplified" License
22 stars 7 forks source link

Can I use this code for evaluating the same dataset after I convert it into 2d images #9

Open sebyo opened 3 years ago

sebyo commented 3 years ago

at first I convert the images to png to use them for the segmentation process and then I convert my predicted images from png to mhd and used them as input for the algorithm provided But it didn't work any suggetions !

catactg commented 3 years ago

Hello Could you provide more details on how "it didn't work"? Did you get an error? Did you get no metrics output? I suggest you visualize the MHD files first to make sure they look correct. For example, using ITK-SNAP (or Paraview, 3DSlicer) you can load both the ground-truth MHD files and your predicted MHD files and make sure segmented LA overlap.

sebyo commented 3 years ago

first I took the 3d images for exemple : a001 I convert it in axial ,sagittal, coronal plan .So for one 3d image I got 750 2d images .In the end from the 10 images provided I got 7500 2d images then I used them to train my model. and I did the same for th test images I convert them and got 13500 2d images after that I predicted the test images using my model,as a result I got the generated labels Now for the evaluation the images should be .mhd so I did convert them bach to mhd both the ground truth and label but when I took an exemple of my generated label and the ground truth label the algorithm is unable to read them ? my thoughts is because of the standardisation procees and the files attatched to each raw image ?