Adaptive Smoothing in fMRI Data Processing Neural Networks

TL;DR

This paper introduces an adaptive smoothing layer in deep learning models for fMRI data, enabling dynamic adjustment of smoothing levels to improve brain activity analysis.

Contribution

It proposes a novel adaptive smoothing layer within end-to-end neural networks for fMRI, enhancing data processing by optimizing smoothing per brain volume.

Findings

Adaptive smoothing improves task classification accuracy.

The method outperforms fixed smoothing approaches.

Demonstrated on real fMRI data with finger tapping tasks.

Abstract

Functional Magnetic Resonance Imaging (fMRI) relies on multi-step data processing pipelines to accurately determine brain activity; among them, the crucial step of spatial smoothing. These pipelines are commonly suboptimal, given the local optimisation strategy they use, treating each step in isolation. With the advent of new tools for deep learning, recent work has proposed to turn these pipelines into end-to-end learning networks. This change of paradigm offers new avenues to improvement as it allows for a global optimisation. The current work aims at benefitting from this paradigm shift by defining a smoothing step as a layer in these networks able to adaptively modulate the degree of smoothing required by each brain volume to better accomplish a given data analysis task. The viability is evaluated on real fMRI data where subjects did alternate between left and right finger tapping tasks.