Hi there, We’re a group of undergrads working on Task R1 and still unfamiliar about reconstruction image. We saw that P007 isn’t in the training data. Should we train one model on all cases (P001–P061) and then feed it P007’s undersampled k-space, or is there another way you expect us to handle P007? Also, each training patient has many mask files (for example, cine_lax_3ch_mask_ktGaussian8/16/24, cine_lax_3ch_mask_ktRadial8/16/24, cine_lax_4ch_mask_ktRadial8/16/24, etc.), but for validation and submission we only need three specific ones: cine_lax_3ch_kus_Uniform8, cine_lax_4ch_kus_ktRadial16, and cine_sax_kus_ktGaussian24. So, which files should we use as inputs when training our model? Thanks a lot for your help!

Created by Cappi Wong cappiw7
When preparing the challenge dataset, we randomly divide the entire internal dataset (excluding external datasets) into training, validation, and test sets. As a result, you may notice some gaps in the numbering. However, this does not affect your training process. We recommend using all the data we provide for training. The sampling mask is provided to assist participants in training their models, but you are also free to construct your own sampling files. Ultimately, participants are required to complete validation and testing on the undersampled datasets that we provide.

Questions About Inputs and Masks for Regular Task R1 page is loading…