Automatic classification of sleep stages using EEG signals and convolutional neural networks

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(1), P. e0297582 - e0297582

Published: Jan. 26, 2024

Sleep stages classification is one of the new topics in studying human life quality because it plays a crucial role getting healthy lifestyle. Abnormal changes or absence normal sleep may lead to different diseases such as heart-related diseases, diabetes, and obesity. In general, staging analysis can be performed using electroencephalography (EEG) signals. This study proposes convolutional neural network (CNN) based methodology for stage EEG signals taken by six channels transformed into time-frequency images. The proposed consists three major steps: (i) segment signal epochs with 30 seconds length, (ii) convert 2D representation analysis, (iii) feed CNN. results showed that robust achieved very high accuracy 99.39% channel C4-A1. All other have values above 98.5%, which indicates any used accuracy. outperformed methods literature terms overall single It expected provide great benefit physicians, especially neurologists; providing them powerful tool support clinical diagnosis sleep-related diseases.

Language: Английский

---

Validation and interpretation of a multimodal drowsiness detection system using explainable machine learning

Computer Methods and Programs in Biomedicine, Journal Year: 2023, Volume and Issue: 243, P. 107925 - 107925

Published: Nov. 8, 2023

Drowsiness behind the wheel is a major road safety issue with efforts focused on developing drowsy driving detection systems. However, most studies using physiological signals have 'black box' machine learning classifier, much less focus 'robustness' and 'explainability'-two crucial properties of trustworthy model. Therefore, this study has multiple validation techniques to evaluate overall performance such system supervised learning-based classifiers then unbox black box model explainable learning.Driving was simulated via 30-minute psychomotor vigilance task while participants reported their level subjective sleepiness signals: electroencephalogram (EEG), electrooculogram (EOG) electrocardiogram (ECG) being recorded. Six different techniques, comprising subject-dependent independent were applied for robustness testing three classifiers, namely K-nearest neighbours (KNN), support vector machines (SVM) random forest (RF), two methods, SHapley Additive exPlanation (SHAP) analysis partial dependency (PDA) leveraged interpretation.The identified leave one participant out, subject-independent technique be useful, best sensitivity 70.3 %, specificity 82.2 an accuracy 80.1 % classifier in addressing autocorrelation due inter-individual differences signals. Moreover, results suggest important features drowsiness detection, clear cut-off decision boundary.The implication will ensure rigorous approach enhancing safety. The show promise real-life deployment physiological-signal based in-vehicle system, higher reliability explainability, along lower cost.

Language: Английский

---

Rethinking Delayed Hemodynamic Responses for fNIRS Classification

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Journal Year: 2023, Volume and Issue: 31, P. 4528 - 4538

Published: Jan. 1, 2023

Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technology for monitoring cerebral hemodynamic responses. Enhancing fNIRS classification can improve the performance of brain-computer interfaces (BCIs). Currently, deep neural networks (DNNs) do not consider inherent delayed responses signals, which causes many optimization and application problems. Considering kernel size receptive field convolutions, as domain knowledge are introduced into classification, concise efficient model named fNIRSNet proposed. We empirically summarize three design guidelines fNIRSNet. In subject-specific subject-independent experiments, outperforms other DNNs on open-access datasets. Specifically, with only 498 parameters 6.58% higher than convolutional network (CNN) millions mental arithmetic tasks floating-point operations (FLOPs) much lower CNN. Therefore, friendly to practical applications reduces hardware cost BCI systems. It may inspire more research knowledge-driven models BCIs. Code available at https://github.com/wzhlearning/fNIRSNet.

Language: Английский

---

An Improved Microaneurysm Detection Model Based on SwinIR and YOLOv8

Bioengineering, Journal Year: 2023, Volume and Issue: 10(12), P. 1405 - 1405

Published: Dec. 8, 2023

Diabetic retinopathy (DR) is a microvascular complication of diabetes. Microaneurysms (MAs) are often observed in the retinal vessels diabetic patients and represent one earliest signs DR. Accurate efficient detection MAs crucial for diagnosis In this study, an automatic model (MA-YOLO) proposed MA fluorescein angiography (FFA) images. To obtain detailed features improve discriminability FFA images, SwinIR was utilized to reconstruct super-resolution solve problems missed small feature information loss, layer added between neck head sections YOLOv8. enhance generalization ability MA-YOLO model, transfer learning conducted high-resolution images low-resolution avoid excessive penalization due geometric factors address sample distribution imbalance, loss function optimized by taking Wise-IoU as bounding box regression loss. The performance compared with that other state-of-the-art models, including SSD, RetinaNet, YOLOv5, YOLOX, YOLOv7. results showed had best detection, shown its optimal metrics, recall, precision, F1 score, AP, which were 88.23%, 97.98%, 92.85%, 94.62%, respectively. Collectively, suitable can assist ophthalmologists progression

Language: Английский

---

A systematic review and research recommendations on artificial intelligence for automated cervical cancer detection

Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery, Journal Year: 2024, Volume and Issue: 14(6)

Published: July 15, 2024

Abstract Early diagnosis of abnormal cervical cells enhances the chance prompt treatment for cancer (CrC). Artificial intelligence (AI)‐assisted decision support systems detecting are developed because manual identification needs trained healthcare professionals, and can be difficult, time‐consuming, error‐prone. The purpose this study is to present a comprehensive review AI technologies used pre‐cancerous lesions cancer. includes studies where was applied Pap Smear test (cytological test), colposcopy, sociodemographic data other risk factors, histopathological analyses, magnetic resonance imaging‐, computed tomography‐, positron emission tomography‐scan‐based imaging modalities. We performed searches on Web Science, Medline, Scopus, Inspec. preferred reporting items systematic reviews meta‐analysis guidelines were search, screen, analyze articles. primary search resulted in identifying 9745 followed strict inclusion exclusion criteria, which include windows last decade, journal articles, machine/deep learning‐based methods. A total 58 have been included further analysis after identification, screening, eligibility evaluation. Our shows that deep learning models techniques, whereas machine data. convolutional neural network‐based features yielded representative characteristics CrC. also highlights need generating new easily accessible diverse datasets develop versatile CrC detection. model explainability uncertainty quantification increase trust clinicians stakeholders decision‐making automated detection models. suggests privacy concerns adaptability crucial deployment hence, federated meta‐learning should explored. This article categorized under: Fundamental Concepts Data Knowledge > Explainable Technologies Machine Learning Classification

Language: Английский

---