How to fast - track the early identification of pre-school autistic children? By comparing diverse machine learning methods with resting - state fNIRS features / Yaou ZHAO in Research in Autism, 138 (October 2026)  

Public ISBD [article]  inResearch in Autism > 138 (October 2026) . - p.203021 Titre : How to fast - track the early identification of pre-school autistic children? By comparing diverse machine learning methods with resting - state fNIRS features Type de document : texte imprimé Auteurs : Yaou ZHAO, Auteur ; Kaiyun LI, Auteur ; Yuehui CHEN, Auteur ; Bang DU, Auteur ; Liu CHEN, Auteur ; Mingxue WANG, Auteur ; Ziyang QI, Auteur ; Yi CAO, Auteur ; Ruizhi HAN, Auteur ; Xiaotong LI, Auteur ; Wuyue WANG, Auteur ; Qingfang MENG, Auteur ; Fanlu JIA, Auteur ; Jiayi ZHANG, Auteur ; Xinyu LIU, Auteur ; Juan FU, Auteur Article en page(s) : p.203021 Langues : Anglais (eng) Mots-clés : Autism  Rs-fNIRS  Machine learning  Feature selection  The change in the Binary Entropy Index. décimale : PER Périodiques Résumé : Functional near-infrared spectroscopy (fNIRS) combined with machine learning (ML) has been increasingly applied to distinguish autism spectrum disorder (ASD) based on resting-state (rs) hemodynamic fluctuations. This study constructed rs-fNIRS datasets comprised of multi-2-minute oxy-hemoglobin (HbO), deoxy-hemoglobin (Hb), and total hemoglobin (HbT) signals (two- or four-times collection for each child) from 50 fNIRS channels. Four evolutionary computation channel selection algorithms (GA, DE, PSO, PBIL) and 12 classifiers (KNN, RC, NB, DT, ET, LSVM, RBF-SVM, RF, ADA, XGB, E-KNN, E-SVM) were firstly employed to test 12 features (mean, std, kurtosis, skewness, BinEn, ApEN, Δmean, Δstd, Δkurtosis, Δskewness, ΔBinEn, ΔApEN) of the HbO,Hb and HbT. The 3 s time series curve closely aligns with the 2-minute original data, and ΔBinEn being the optimal discriminative feature. The combination of E-SVM and E-KNN ensemble classifiers with the PBIL algorithm achieved the best performance across HbO, Hb, and HbT data. Validation experiments demonstrated the method's robustness, with an average classification accuracy exceeding 90% and identified key contribution channels (CH1, CH11, CH20, CH24, CH27, CH33, CH42). These findings revealed that rs-fNIRS-ML can serve as an efficient clinical tool for early ASD diagnosis in preschoolers. En ligne : https://dx.doi.org/https://doi.org/10.1016/j.reia.2026.203021 Permalink : https://www.cra-rhone-alpes.org/cid/opac_css/index.php?lvl=notice_display&id=594 [article] How to fast - track the early identification of pre-school autistic children? By comparing diverse machine learning methods with resting - state fNIRS features [texte imprimé] / Yaou ZHAO, Auteur ; Kaiyun LI, Auteur ; Yuehui CHEN, Auteur ; Bang DU, Auteur ; Liu CHEN, Auteur ; Mingxue WANG, Auteur ; Ziyang QI, Auteur ; Yi CAO, Auteur ; Ruizhi HAN, Auteur ; Xiaotong LI, Auteur ; Wuyue WANG, Auteur ; Qingfang MENG, Auteur ; Fanlu JIA, Auteur ; Jiayi ZHANG, Auteur ; Xinyu LIU, Auteur ; Juan FU, Auteur . - p.203021. Langues : Anglais (eng) in Research in Autism > 138 (October 2026) . - p.203021 Mots-clés : Autism  Rs-fNIRS  Machine learning  Feature selection  The change in the Binary Entropy Index. décimale : PER Périodiques Résumé : Functional near-infrared spectroscopy (fNIRS) combined with machine learning (ML) has been increasingly applied to distinguish autism spectrum disorder (ASD) based on resting-state (rs) hemodynamic fluctuations. This study constructed rs-fNIRS datasets comprised of multi-2-minute oxy-hemoglobin (HbO), deoxy-hemoglobin (Hb), and total hemoglobin (HbT) signals (two- or four-times collection for each child) from 50 fNIRS channels. Four evolutionary computation channel selection algorithms (GA, DE, PSO, PBIL) and 12 classifiers (KNN, RC, NB, DT, ET, LSVM, RBF-SVM, RF, ADA, XGB, E-KNN, E-SVM) were firstly employed to test 12 features (mean, std, kurtosis, skewness, BinEn, ApEN, Δmean, Δstd, Δkurtosis, Δskewness, ΔBinEn, ΔApEN) of the HbO,Hb and HbT. The 3 s time series curve closely aligns with the 2-minute original data, and ΔBinEn being the optimal discriminative feature. The combination of E-SVM and E-KNN ensemble classifiers with the PBIL algorithm achieved the best performance across HbO, Hb, and HbT data. Validation experiments demonstrated the method's robustness, with an average classification accuracy exceeding 90% and identified key contribution channels (CH1, CH11, CH20, CH24, CH27, CH33, CH42). These findings revealed that rs-fNIRS-ML can serve as an efficient clinical tool for early ASD diagnosis in preschoolers. En ligne : https://dx.doi.org/https://doi.org/10.1016/j.reia.2026.203021 Permalink : https://www.cra-rhone-alpes.org/cid/opac_css/index.php?lvl=notice_display&id=594

---

1 (1 - 1 / 1) Par page : 25 50 100 200