xuchen312
commented
4 years ago The title and abstract are correct. But could you please update the author list by adding the following two coauthors? Thank you very much!
- Wouter Weeda, Leiden University, Leiden, Netherlands (@wdweeda)
- Jelle Goeman, Leiden University Medical Center, Leiden, Netherlands
jsheunis
ARIbrain: Interactive clustering with spatial activation guarantees using all-resolutions inference
By: Xu Chen, Leiden University Medical Center & Leiden University, Leiden, Netherlands Wouter Weeda, Leiden University, Leiden, Netherlands (@wdweeda) Jelle Goeman, Leiden University Medical Center, Leiden, Netherlands
Abstract
Cluster inference approaches have become standard in fMRI data analysis nowadays, however, they are commonly criticized for lack of spatial specificity. The widely used parametric cluster inference, based on random field theory (RFT), has been reported to induce invalid results with highly inflated false positive rate (FPR) (Eklund et al., 2016). Our recently proposed all-resolutions inference (ARI; Rosenblatt et al., 2018) is a valid cluster-wise inference approach using closed testing with strongly controlled family-wise error rate (FWER), which has also been shown to have comparable detection sensitivity to RFT-based methods (Weeda et al., 2019). Although ARI allows completely free choice of clusters and returns the lower bound for the proportion of truly activated voxels within the selected clusters, it does not suggest any specific clusters without exploring all possible regions of the brain. Localization of activation thus requires users to drill down into clusters to obtain a sufficient TDP.
Therefore we now introduce a highly efficient adaptive clustering algorithm using ARI to alleviate the spatial specificity problem by finding the largest contiguous clusters with spatial activation guarantees using a user-chosen threshold of true discovery proportion (TDP). In contrast to the conventional clusters defined using a single cluster-forming threshold, the resulting clusters by this algorithm are essentially supra-threshold clusters created by applying different cluster-forming thresholds automatically, and each cluster with the TDP exceeding the pre-specified threshold. This algorithm also enables the use of arbitrary TDP thresholds at different locations in the brain to provide users with more flexibility. For each algorithm-generated cluster, users can interactively change the cluster size or the TDP by enlarging the cluster for a decreased TDP or reducing the cluster for an increased TDP. Under these circumstances, the FWER is always under full control due to the simultaneity of ARI.
We are currently building an interactive shiny web app “ARIbrain” to offer an easy implementation of the algorithm, and we are also working on the development of toolboxes including SPM extension and Python module to facilitate and support the application of this algorithm in real data analysis.
[1] Eklund A., Nichols T.E., Knutsson H. (2016). Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates. PNAS 113(28):7900-7905. [2] Rosenblatt J.D., Finos l., Weeda W.D., Solari A., Goeman J.J. (2018). All-Resolutions Inference for brain imaging. Nueroimage 181:786-796. [3] Weeda W., van Kempen M., Chen X., Rosenblatt J., Finos L., Solari A., Goeman J. (2019). The ARIbrain App: A versatile application for fMRI data analysis based on valid circular inference (Poster). OHBM 2019, Rome, Italy.
Useful Links
https://github.com/wdweeda/ARIbrain-app []()
Tagging @xuchen312