search

Project-MONAI / MONAILabel

MONAI Label is an intelligent open source image labeling and learning tool.
https://docs.monai.io/projects/label
Apache License 2.0
595 stars 191 forks source link

Only one channel is recognized out of 15 channels #1652

Open mengab1 opened 6 months ago

mengab1 commented 6 months ago

[2024-03-10 14:28:38,126] [18300] [MainThread] [INFO] (main:61) - Result: {"rank": 0, "current_epoch": 2000, "current_iteration": 24000, "total_epochs": 2000, "total_iterations": 12, "epoch": 2000, "start_ts": 1710020807, "total_time": "8:41:50", "best_metric": 0.787301242351532, "train": {"metrics": {"train_mean_dice": 0.8769383430480957, "train_nodule_mean_dice": 0.0, "train_bone_mean_dice": 0.0, "train_aorta_mean_dice": 0.0, "train_pulmonary_artery_mean_dice": 0.0, "train_pulmonary_vein_mean_dice": 0.0, "train_vein_mean_dice": 0.0, "train_trachea_mean_dice": 0.0, "train_esophagus_mean_dice": 0.0, "train_heart_mean_dice": 0.0, "train_lung_upper_lobe_left_mean_dice": 0.0, "train_lung_lower_lobe_left_mean_dice": 0.0, "train_lung_upper_lobe_right_mean_dice": 0.0, "train_lung_middle_lobe_right_mean_dice": 0.0, "train_lung_lower_lobe_right_mean_dice": 0.0, "train_skin_mean_dice": 0.9323155283927917}, "key_metric_name": "train_mean_dice", "best_metric": 0.9949172139167786, "best_metric_epoch": 1851}, "eval": {"metrics": {"val_mean_dice": 0.7803362011909485, "val_nodule_mean_dice": 0.0, "val_bone_mean_dice": 0.0, "val_aorta_mean_dice": 0.0, "val_pulmonary_artery_mean_dice": 0.0, "val_pulmonary_vein_mean_dice": 0.0, "val_vein_mean_dice": 0.0, "val_trachea_mean_dice": 0.0, "val_esophagus_mean_dice": 0.0, "val_heart_mean_dice": 0.0, "val_lung_upper_lobe_left_mean_dice": 0.0, "val_lung_lower_lobe_left_mean_dice": 0.0, "val_lung_upper_lobe_right_mean_dice": 0.0, "val_lung_middle_lobe_right_mean_dice": 0.0, "val_lung_lower_lobe_right_mean_dice": 0.0, "val_skin_mean_dice": 0.9444830417633057}, "key_metric_name": "val_mean_dice", "best_metric": 0.787301242351532, "best_metric_epoch": 1669}} [2024-03-10 14:28:39,100] [13122] [ThreadPoolExecutor-1_0] [INFO] (monailabel.utils.async_tasks.utils:83) - Return code: 0

截屏2024-03-10 21 19 27 截屏2024-03-10 21 25 17 截屏2024-03-10 21 26 50
diazandr3s commented 6 months ago

Hi @mengab1,

Thanks for posting this. I was wondering whether the dataset has the label indices you show in the self.labels dictionary. Can you check if the label files to train the model have those indices consistently throughout the dataset?

Let us know,

mengab1 commented 6 months ago

How to set self.labels correctly I start training from scratch

diazandr3s commented 6 months ago

Hi @mengab1,

Are you trying to segment all these regions from scratch? If yes, how many volumes did you segment before training the model? Just to let you know, you could also use the MONAIAuto3DSeg extension in Slicer to bootstrap your annotation workflow. There is a model called mediastinal_anatomy you may find useful

Just download the preview release from Slicer (https://download.slicer.org/) and then install MONAIAuto3DSeg via the Extension Manager. With the predictions you get from that model, you can create ground truth to train and use the active learning strategies from MONAI Label.

Hope this helps,