fMRI alignment benchmarking on the IBC dataset

📄 View report

Functional Brain Alignment is key to understanding individual idiosyncracies in brain functional disposition and is used in a variety of experimental setups such as brain decoding. There are many ways to do brain alignment, that fall into two broad categories,pairwise alignmentand template alignment. There are many ways to perform template alignment, such as the Shared Response Model (Richard et al., 2019), developed at INRIA Saclay. We are studying the Individual Neural Tuning model (Feilong et al., 2022) from HaxbyLab. It aims at doing template alignment through hyperalignment, a method that uses searchlight decomposition of brains to do piecewise local alignments and combine the results. Interesting claims are made in the paper regarding correlations, with impressive results such as Intra- Subjects correlations reaching 0.4 in some setups. Those numbers caught our attention and are the main reason of this analysis. We aim to replicate some results from (Feilong et al., 2022), and further test the model on more data such as contrasts from the IBC Dataset. We will also compare the performances to other similar alignment techniques such as FastSRM. Our objectives are the following :

• Verify the validity of the model and the claims made.
• Benchmark the model on higher resolution data, and with more timeframes.
• Use other benchmarks.