Hi everyone, I would like to train a model to segment multiple sclerosis plaques on 3D FLAIR images, one by one, each with distinct labels. I’ve started with a dataset of 10 patients just to get started before tackling segmentation on a larger number of cases. As described in the tutorial “MONAI Label - Training from Scratch (https://www.youtube.com/watch?v=3HTh2dqZqew)”, I modified the dictionary in the deepedit.py file by adding labels like “lesion1, lesion2, lesion3,” etc. (I’ve included 7 labels) and then changed “use_pretrained_model” to “false”. As seen in the screenshot, I’ve just reached an accuracy of 27%. However, even after adding additional labels for lesions in the segment editor (such as lesion8, lesion9, lesion10, etc., since some cases have more than 7 plaques), it seems that lesions beyond the seventh one are not considered during training (see the terminal screenshot). How can I resolve these issues? Would training on more cases lead to better accuracy?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Training Segmentation Model with Custom Labels in MonaILabel | 2 | 620 | October 11, 2023 | |
MONAILABEL plugin | pelvis segmentation in MRI | 8 | 972 | March 24, 2022 | |
Air segmentation using MONAILabel_improve accuracy | 0 | 188 | May 15, 2023 | |
RunTimeError_Multilabel | 0 | 162 | June 28, 2023 | |
MonaiLabel Models; Training from Scratch multiple times | 0 | 89 | June 24, 2024 |