AI aided annotation of NSCLC-Radiomics, LIDC-IDRI, Duke-Breast-Cancer-MRI, I-SPY2 Trial

Support Data
1

We are looking at AI aided annotation of 4 datasets -
NSCLC-Radiomics, LIDC-IDRI, Duke-Breast-Cancer-MRI, I-SPY2 Trial

  1. We are considering three models -
    UNETR for NSCLC-Radiomics and Duke-Breast-Cancer-MRI
    Bi-Directional ConvLSTM U-Net with Densely Connected Convolutions. for LIDC-IDRI
    Detectron2 from Meta AI for I-SPY2 Trial

(UNETR - Transformers for 3D Medical Image Segmentation - [2103.10504] UNETR: Transformers for 3D Medical Image Segmentation)

(Detectron2 - Google Colab)

Would you recommend other models?

  1. For QC metrics we are considering IOU, Dice coefficient and crossentropy loss. Would you recommend other metrics?

Warm regards,
Aparna

2

@Aparna_Prabhu we are here to answer your questions about IDC and support your use of data from IDC for your work. Your questions go beyond the scope of responsibilities of the IDC team.

For selecting models and evaluation metrics my recommendation would be to consult the latest developments in the literature for the specific annotation tasks. You could also reach out to the submitters of the collections you mentioned and ask for their opinion. If there are other users within the IDC community who might have recommendations, they are welcomed to respond in this thread.

1 Like