Back

Cross-validation coordinate analysis (CVCA): performing coordinate based meta-analysis using cross-validation

Tench, C. R.

2024-03-15 radiology and imaging
10.1101/2024.03.13.24304212 medRxiv
Show abstract

Coordinate based meta-analysis (CBMA) can be used to estimate where a future neuroimaging study testing a particular hypothesis might report summary results (activation foci, for example). However, current methods cannot be validated for all possible analyses, and use empirical features that might not always be ideal. Indeed, the various algorithms that perform CBMA tend to produce somewhat different results even on the same data. Furthermore, the use of null hypothesis significance testing (NHST) in the algorithms is not strictly in keeping with meta-analysis, where the aim is usually effect estimation rather than statistical significance. Given the limitations of current CBMA algorithms, a new algorithm has been developed that will perform its analysis using cross validation: cross-validation coordinate analysis (CVCA). Although a full validation cannot be performed, CVCA optimises its parameters in such a way as to make the analysis results most similar to the held-out data. The algorithm can be used as a stand-alone meta-analysis method, or to provide confidence in results from other algorithms where no validation is performed. Software to perform CVCA is freely available.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.