Simple But Effective: Rethinking the Ability of Deep Learning in fNIRS to Exclude Abnormal Input

TL;DR

This work proposes integrating metric learning and supervised methods into fNIRS research to improve networks capability in identifying and excluding out-of-distribution outliers, and significantly enhanced the performance of various networks in fNIRS, particularly transformer-based one which shows the great improvement in reliability.

Abstract

Functional near-infrared spectroscopy (fNIRS) is a non-invasive technique for monitoring brain activity. To better understand the brain, researchers often use deep learning to address the classification challenges of fNIRS data. Our study shows that while current networks in fNIRS are highly accurate for predictions within their training distribution, they falter at identifying and excluding abnormal data which is out-of-distribution, affecting their reliability. We propose integrating metric learning and supervised methods into fNIRS research to improve networks capability in identifying and excluding out-of-distribution outliers. This method is simple yet effective. In our experiments, it significantly enhances the performance of various networks in fNIRS, particularly transformer-based one, which shows the great improvement in reliability. We will make our experiment data available on GitHub.

Figures (9)

• Figure 2: In each cross-validation, inputing the in-distribution data or out-of-distribution data (e.g. meaningless noise), and observing the confidence of the output after the SoftMax. Surprisingly, obtaining high confidence under normal data and noise input conditions, which shows the network's inability to exclude meaningless noise input.
• Figure 3: (a) Sensor location layout for MA shin2016open. (b) Sensor location layout for UFFT bak2019open.
• Figure 4: The structure of neural network.
• Figure 5: (a) Explicit expression of cosine distance function. (b) Explicit expression of Euclidean distance function.
• Figure 6: Visualization of the results of fNIRS-Transformer as feature extraction on the detector feature subspace.