Deep learning in fNIRS: a review

Eastmond, Condell; Subedi, Aseem; De, Suvranu; Intes, Xavier

doi:10.1117/1.nph.9.4.041411

Cited by 51 publications

(42 citation statements)

References 106 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For this compile stage, we set the learning rate, batch size, and the number of epochs as HPs for auto-tuning. Unlike most works that chose typical HPs like the number of neurons, dropout rate, learning rates, and the number of epochs for auto-tuning [29], [44], [45], we also included activation functions, batch size, and most importantly, network structures as our auto-tuned HPs, which were as much we can auto-tune as possible. The 2D and 3D CNN networks were constructed similarly to the MLP network.…”

Section: Model Constructionmentioning

confidence: 99%

“…Our research found three main challenges we wanted to solve using fNIRS signals and deep learning methods to classify MCI vs. CN. Firstly, while many works used fNIRS signals with deep learning methods for other diagnoses (like detecting the pain intensity, epileptic seizure, and autism spectrum disorder), cortical analysis, and brain-computer interface and showed accurate classification results [27]- [29], only a few works utilized deep learning methods to detect MCI vs. CN [26], [30]. Secondly, deep learning methods require a large amount of data to prevent overfitting, yet the existing studies in our area mostly had small datasets of less than 50 participants [29].…”

Section: Introductionmentioning

confidence: 99%

“…Firstly, while many works used fNIRS signals with deep learning methods for other diagnoses (like detecting the pain intensity, epileptic seizure, and autism spectrum disorder), cortical analysis, and brain-computer interface and showed accurate classification results [27]- [29], only a few works utilized deep learning methods to detect MCI vs. CN [26], [30]. Secondly, deep learning methods require a large amount of data to prevent overfitting, yet the existing studies in our area mostly had small datasets of less than 50 participants [29]. Finally, few studies have investigated how multi-dimensional fNIRS features affect classification results, despite machine learning and deep learning having proven effective at diagnosing MCI.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Comparing Multi-Dimensional fNIRS Features Using Bayesian Optimization-Based Neural Networks for Mild Cognitive Impairment (MCI) Detection

Zhang

Yang

Fan

et al. 2023

IEEE Trans. Neural Syst. Rehabil. Eng.

View full text Add to dashboard Cite

The diagnosis of mild cognitive impairment (MCI), a prodromal stage of Alzheimer's disease (AD), is essential for initiating timely treatment to delay the onset of AD. Previous studies have shown the potential of functional near-infrared spectroscopy (fNIRS) for diagnosing MCI. However, preprocessing fNIRS measurements requires extensive experience to identify poor-quality segments. Moreover, few studies have explored how proper multidimensional fNIRS features influence the classification results of the disease. Thus, this study outlined a streamlined fNIRS preprocessing method to process fNIRS measurements and compared multi-dimensional fNIRS features with neural networks in order to explore how temporal and spatial factors affect the classification of MCI and cognitive normality. More specifically, this study proposed using Bayesian optimization-based auto hyperparameter tuning neural networks to evaluate 1D channel-wise, 2D spatial, and 3D spatiotemporal features of fNIRS measurements for detecting MCI patients. The highest test accuracies of 70.83%, 76.92%, and 80.77% were achieved for 1D, 2D, and 3D features, respectively. Through extensive comparisons, the 3D time-point oxyhemoglobin feature was proven to be a more promising fNIRS feature for detecting MCI by using an fNIRS dataset of 127 participants. Furthermore, this study presented a potential approach for fNIRS data processing, and the designed models required no manual hyperparameter tuning, which promoted the general utilization of fNIRS modality with neural network-based classification to detect MCI.

show abstract

Section: Model Constructionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Comparing Multi-Dimensional fNIRS Features Using Bayesian Optimization-Based Neural Networks for Mild Cognitive Impairment (MCI) Detection

Zhang

Yang

Fan

et al. 2023

IEEE Trans. Neural Syst. Rehabil. Eng.

View full text Add to dashboard Cite

show abstract

“…Therefore, performing SSH and TTL on LI 11 is a good choice for ARRMs neuroimaging research. In recent years, functional near-infrared spectroscopy (fNIRS) has emerged as a well-established imaging tool for neuroscience research (Eastmond et al, 2022). It has a high temporal resolution (Tak and Ye, 2014) and the capacity for monitoring in real clinical settings (Dybvik and Steinert, 2021;Gossé et al, 2022).…”

Section: Introductionmentioning

confidence: 99%

Methodological issues of the central mechanism of two classic acupuncture manipulations based on fNIRS: suggestions for a pilot study

Cao²,

Chen³

et al. 2023

Front. Hum. Neurosci.

View full text Add to dashboard Cite

Background: Acupuncture reinforcing-reducing manipulation (ARRM) is a necessary procedure of traditional Chinese acupuncture and an essential factor affecting the therapeutic effect of acupuncture. Shaoshanhuo reinforcing method (SSH) and Toutianliang reducing method (TTL) are the most representative ARRMs. They integrate six single ARRMs and pose distinguished therapeutic effects of acupuncture. However, due to the complexity, diversity, and variation, investigating the mechanism of these two classic manipulations is insufficient. The neuroimaging technique is an important method to explore the central mechanism of SSH and TTL. This study attempted to design a randomized crossover trial based on functional near-infrared spectroscopy (fNIRS) to explore the mechanism of SSH and TTL, meanwhile, provide valuable methodological references for future studies.Methods: A total of 30 healthy subjects were finally included and analyzed in this study. fNIRS examination was performed to record the neural responses during the two most representative ARRMs. The cortical activation and the inter-network functional connectivity (FC) were explored.Results: The results found that SSH and TTL could elicit significant cerebral responses, respectively, but there was no difference between them.Conclusion: Neuroimaging techniques with a higher spatiotemporal resolution, combinations of therapeutic effects, and strict quality control are important to neuroimaging studies on SSH and TTL.

show abstract

“…To our best knowledge, this is the first review that covers machine learning studies focusing on biomarker research using fNIRS. There is a recent review focusing on deep learning (Eastmond et al, 2022), however, as we stated above we also discussed the regions and potential biomarkers used as a feature that provides high diagnostic accuracy.…”

Section: Introductionmentioning

confidence: 99%

Diagnostic Machine Learning Applications on Clinical Populations using Functional Near Infrared Spectroscopy: A Review

Eken

Nassehi

Eroğul

2023

Preprint

View full text Add to dashboard Cite

Objective : There is a growing interest in functional near-infrared spectroscopy (fNIRS) among researchers due to its remarkable advantages such as ease of use, being inexpensive and less tolerance to motion artifacts compared to other neuroimaging modalities. Also, its interaction with machine learning (ML) approaches was inevitable like other neuroimaging modalities for different purposes such as diagnostic classification of diseases or prediction of disease severity due to the lack of robust and objective biomarkers. A review of literature is carried out to understand the evolution of biomarker research on clinical populations and clinical states by combining fNIRS and ML. Approach : In this review, article search was carried out in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) standard and 62 studies were evaluated and we provided a general overview by using fNIRS, particularly in clinical populations and some clinically relevant conditions on healthy populations. Also, potential biomarkers that were found in these studies and some popular ML algorithms that had been used for the prediction or classification of fNIRS data were discussed. Main Results : There is an increasing trend to perform ML applications on fNIRS data on biomarker research related to different clinical fields. Among these studies, few were able to have a notable number of participants for classification and clinical state prediction. Oxy-hemoglobin was used more than deoxy-hemoglobin in ML-based studies as a potential feature source. Significance : Using ML on fNIRS data might be a promising approach to revealing specific biomarkers for either diagnostic classification of diseases or prediction of clinical conditions.

show abstract

Deep learning in fNIRS: a review

Cited by 51 publications

References 106 publications

Comparing Multi-Dimensional fNIRS Features Using Bayesian Optimization-Based Neural Networks for Mild Cognitive Impairment (MCI) Detection

Comparing Multi-Dimensional fNIRS Features Using Bayesian Optimization-Based Neural Networks for Mild Cognitive Impairment (MCI) Detection

Methodological issues of the central mechanism of two classic acupuncture manipulations based on fNIRS: suggestions for a pilot study

Diagnostic Machine Learning Applications on Clinical Populations using Functional Near Infrared Spectroscopy: A Review

Contact Info

Product

Resources

About