RANO FL Study

This is the documentation for the Federated Learning Study conducted by RANO in collaboration with MLCommons.

Home
Application Setup
Process the Data
Run the Application
Extras
Running Federated Training
ITCR Connectivity

Federated Learning for Postoperative Segmentation of Treated glioblastoma (FL-PoST)

Announcement

Duke University, Indiana University, and the Response Assessment in Neuro-Oncology (RANO) team are excited to announce a new collaborative study seeking to extend the FeTS initiative to the post-operative/post-treatment setting using a large, diverse imaging dataset derived from patients with previously treated glioblastoma, which is hereby termed Federated Learning for Postoperative Segmentation of Treated glioblastoma (FL-PoST) or simply FeTS 2.0.

The FL-PoST initiative aims to revolutionize the quantification of brain tumors by establishing an international federation of healthcare institutions and developing an open-source software ecosystem that offers simple, automated, and accurate tumor segmentation free from inter-reader variability.

If you have previously been a part of the preoperative FeTS initiative, thank you for your continued interest and support! We hope that you will consider participating FL-PoST as well.

Purpose of the study

How to participate

  1. Identify a cohort of brain tumor MRI scans from your institution with a confirmed initial diagnosis of glioblastoma (this is the most important step and can start as soon as possible!).
  2. Identify relevant series (T1, T1+Gd, T2, T2-FLAIR) for each subject and their respective post-operative follow-up scans.
  3. Identify a “technical point of contact”, who will conduct the data processing and experiment from your end.
  4. Download and install the FeTS software (detailed instructions along with a new installer is coming soon!).
  5. Preprocess the MRI data using the FeTS software (detailed instructions along with a new installer is coming soon!).
  6. Generate automated segmentations using pre-trained models.
  7. Manually annotate and correct the automated tumor segmentations to generate ground truth labels using the tool of your choice (we recommend ITK-SNAP, CaPTk, or 3D-Slicer). Please ensure that the generated labels are silhouettes (i.e., the entire volume of the tumors are delineated) and not just boundaries.
  8. Run the federated training during the specified 2-week training window (to be announced).

Requirements for participation

Hardware

Software/OS

Dataset

Expected study timeline

Benefits to participating

Where to sign up

If interested, please complete the following survey and a member of the study team will reach out with any next steps: redcap.duke.edu/redcap/surveys/?s=RDXFYPTWA8CDKWFH

More information?

The FL-PoST software is still being finalized. If you sign up to participate in the study, we will send information on installing the software as soon as it is ready. Suggested annotation tools:

If you have any additional questions, please feel free to reach out to: admin [at] fets.ai

People