Procedure¶
The task of this challenge is to automatically detect the location of nodules from volumetric CT images. 888 CT scans from LIDC-IDRI database are provided. To develop a data driven prediction algorithm, the dataset is typically split into training and testing dataset. In order to test the algorithm on all available CT scans, 10-folds cross validation should be used.
The dataset could be downloaded from the 'Data' page. To make the 10-folds cross validation easier to be performed, we have split the dataset into 10 subsets. For fold N:
- split the dataset into a testing and a training dataset. We use the subset N as the testing dataset and the remaining subsets as the training dataset. For the 'false positive reduction' track, testing and training candidates should be extracted on the corresponding testing and training dataset.
- train the algorithm on the training dataset. Tutorial on how to open images or extract candidates is available on the 'Tutorial' page.
- run the trained algorithm on the testing dataset and generate the results file. Check the format of the results file on the 'Evaluation' page.
After running the algorithm on all folds, the results files should be merged. The merged result file can be submitted on the 'Submit' page and will be evaluated by the evaluation script. The result will be displayed on the 'Results' page.
If you have any questions, please email the organizers or post the question on the 'Forum' page.