Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AbsoluteNet: A Deep Learning Neural Network to Classify Cerebral Hemodynamic Responses of Auditory Processing

Behtom Adeli, John Mclinden, Pankaj Pandey, Ming Shao, Yalda Shahriari

TL;DR

This work introduces AbsoluteNet, a dual-stream CNN designed to classify single-trial fNIRS hemodynamic responses to auditory stimuli. By separately extracting spatial-temporal and temporal-spatial features and fusing them through a separable convolution block, and by employing symmetrical activation functions, the model achieves state-of-the-art performance on an auditory oddball dataset, reaching $87.0\%$ accuracy, $84.81\%$ sensitivity, and $89.21\%$ specificity with concatenated $HbO_2$ and $HbR$ signals. An extensive ablation study confirms the importance of dual streams, fusion blocks, and the activation strategy, while a GA-based hyperparameter search further enhances performance. The results highlight the potential of spatio-temporal feature aggregation in fNIRS-based neural decoding and suggest avenues for multimodal extensions and larger-scale validations.

Abstract

In recent years, deep learning (DL) approaches have demonstrated promising results in decoding hemodynamic responses captured by functional near-infrared spectroscopy (fNIRS), particularly in the context of brain-computer interface (BCI) applications. This work introduces AbsoluteNet, a novel deep learning architecture designed to classify auditory event-related responses recorded using fNIRS. The proposed network is built upon principles of spatio-temporal convolution and customized activation functions. Our model was compared against several models, namely fNIRSNET, MDNN, DeepConvNet, and ShallowConvNet. The results showed that AbsoluteNet outperforms existing models, reaching 87.0% accuracy, 84.8% sensitivity, and 89.2% specificity in binary classification, surpassing fNIRSNET, the second-best model, by 3.8% in accuracy. These findings underscore the effectiveness of our proposed deep learning model in decoding hemodynamic responses related to auditory processing and highlight the importance of spatio-temporal feature aggregation and customized activation functions to better fit fNIRS dynamics.

AbsoluteNet: A Deep Learning Neural Network to Classify Cerebral Hemodynamic Responses of Auditory Processing

TL;DR

This work introduces AbsoluteNet, a dual-stream CNN designed to classify single-trial fNIRS hemodynamic responses to auditory stimuli. By separately extracting spatial-temporal and temporal-spatial features and fusing them through a separable convolution block, and by employing symmetrical activation functions, the model achieves state-of-the-art performance on an auditory oddball dataset, reaching accuracy, sensitivity, and specificity with concatenated and signals. An extensive ablation study confirms the importance of dual streams, fusion blocks, and the activation strategy, while a GA-based hyperparameter search further enhances performance. The results highlight the potential of spatio-temporal feature aggregation in fNIRS-based neural decoding and suggest avenues for multimodal extensions and larger-scale validations.

Abstract

In recent years, deep learning (DL) approaches have demonstrated promising results in decoding hemodynamic responses captured by functional near-infrared spectroscopy (fNIRS), particularly in the context of brain-computer interface (BCI) applications. This work introduces AbsoluteNet, a novel deep learning architecture designed to classify auditory event-related responses recorded using fNIRS. The proposed network is built upon principles of spatio-temporal convolution and customized activation functions. Our model was compared against several models, namely fNIRSNET, MDNN, DeepConvNet, and ShallowConvNet. The results showed that AbsoluteNet outperforms existing models, reaching 87.0% accuracy, 84.8% sensitivity, and 89.2% specificity in binary classification, surpassing fNIRSNET, the second-best model, by 3.8% in accuracy. These findings underscore the effectiveness of our proposed deep learning model in decoding hemodynamic responses related to auditory processing and highlight the importance of spatio-temporal feature aggregation and customized activation functions to better fit fNIRS dynamics.

Paper Structure

This paper contains 8 sections, 1 equation, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Experimental Design and Data Preprocessing
  3. Our Proposed Deep Learning Model: AbsoluteNet
  4. Architecture
  5. Training Parameters and Optimization
  6. Ablation Study
  7. Results
  8. Discussion and Conclusion

Figures (4)

  • Figure 1: A) Source-detector placement and corresponding fNIRS channels. B) Auditory oddball paradigm with interleaved deviant (red) and standard (blue) tones and timing sequence.
  • Figure 2: Block diagram of the AbsoluteNet architecture showing temporal-spatial dual streams, separable convolutions, and activation strategy.
  • Figure 3: Performance comparison across multiple deep learning models on the fNIRS dataset.
  • Figure 4: Ablation studies' performance evaluating the impact of removing key components in AbsoluteNet.