TADPOLE test set (D4) now available on ADNI website

[Razvan Marinescu]

Hi everyone,

We received a few requests for accessing the exact test set (D4) used during the final evaluation in June 2019. We have now uploaded it on the ADNI website. It can be found in the same zip file with the rest of the datasets (D1-D3): login to ADNI, follow Download -> Study Data -> Test Data -> Data for Challenges -> "Tadpole Challenge Data". The file name is called TADPOLE_D4_corr.csv (corr means the ventricle volumes were corrected for different Freesurfer versions). Python script TADPOLE_D4.py can be used to generate TADPOLE_D4_corr.csv. Evaluation function evalOneSubmissionD4.py from the github repo can be used to evaluate a final D4 submission (as opposed to a leaderboard submission).

This allows development of subsequent prediction algorithms and evaluation on the same benchmark as the existing 92 TADPOLE forecasts. If you develop improved algorithms, don't forget to upload the predictions on the live D4 leaderboard. 

Regards,

Raz

[dargow...@gmail.com]

sir since my topic is prediction of Alzheimer's diseases and i downloaded MRI image data set from adni. i am facing problem how to use this image data set, sir kindly help me out how can i start working on it

[Razvan Marinescu]

You can predict directly from MRI scans downloaded from ADNI, but you need to pre-process them first and match each scan to the right subject ID and clinical visit.  

If you are ok with using features already extracted from these images, you can start with the standard D1/D2 datasets from TADPOLE_D1_D2.csv. You can download the zip from the ADNI LONI website (see download instructions here https://tadpole.grand-challenge.org/Data/). You can start from the benchmark models we provided here: https://github.com/noxtoby/TADPOLE/tree/master/evaluation. Or the ones from TADPOLE-SHARE: https://tadpole-share.github.io/

Raz