Dear Organizers, Thank you for releasing the weights for the baseline DenseNet model for Task 1. We are looking to run inference on the validation dataset using the weights provided. But unfortunately, the preprocessing details required to provide the correct form of data for inference is not available, and hence we obtain "all zeros" on the input. Could you please provide details on the following: Intensity Normalization used: Z-normalization/(-1,1)/(0,1) ? Input Orientation, Size and Resolution In short, could you provide information on input transformation pipeline? Thank you.

Created by mahesh shakya mahesh_shakya
Dear mahesh_shakya@synapse.org, Thanks for your question. We have provided the training and testing scripts on our GitHub page (https://github.com/LISA2024Challenge). To answer your question, the details can be found in the training folder here: https://github.com/LISA2024Challenge/Task1. We use the following, starting on line 183: train_transforms = Compose( [ LoadImaged(keys=["img"], reader="nibabelreader"), EnsureChannelFirstd(keys=["img"]), NormalizeIntensityd( keys=["img"], nonzero=False, channel_wise=True), CenterSpatialCropd(keys=["img"], roi_size=(x, y, z)), SpatialPadd(keys=["img"],method="symmetric", spatial_size=(x, y, z)), ToTensord(keys=["img"]), ] ) val_transforms = Compose( [ LoadImaged(keys=["img"], reader="nibabelreader"), EnsureChannelFirstd(keys=["img"]), NormalizeIntensityd(keys=["img"], nonzero=False, channel_wise=True), SpatialPadd(keys=["img"],method="symmetric", spatial_size=(x, y, z)), ToTensord(keys=["img"]), ] ) Please let us know if you have any additional questions. Best regards, LISA 2024 Challenge Organizers

Pre-trained baseline Data Preprocessing pipeline for Inference ? page is loading…