Probabilistic ALE

Probablistic ALE: Checking Robustness of Main Effect Results

Here subsampling ist done, systematically repeating an ALE meta-analysis multiple times, leaving out a certain amount of experiments (k) in each iteration. If in the analysis_file the number specified after the P is N - 1 (Overall number of experiments-1), a leave-one-out jackknife analysis is performed, with as many ALE analyses being conducted as there are experiments, each excluding one single experiment. If the number specified after the P is smaller than N - 1, multiple ALE analyses are conducted, each excluding several experiments (k). This approach provides insight into the robustness of the main effect ALE results.

---

Steps in the probabilistic ALE

1. Initial ALE Analysis:

   • Conduct the main effect ALE meta-analysis using all ( N ) experiments to identify significant activation patterns.

2. Iterative Exclusion:

• For each iteration/subsample (default=2500 but the maximum of unique subsamples that can be created from the dataset):
  • Exclude k experiments from the dataset.
  • Perform the ALE meta-analysis on the remaining ( N-k ) experiments.
  • Binarize the resulting thresholded map

3. Create an average map across subsamples:
   • Averaging the binarized maps across all subsample analyses provides a map showing voxelwise relative frequency of significance across subsamples.
   • Larger values indicate more robustness
   • A value of 1 indicates that in all subsample analyses the voxel was significant.

Please note that the resulting map should be interpreted depending on k and that k should be selected based on the absolute number of experiments, i.e. the smaller the dataset, the smaller k should be.

---