Fatigue driving is a major cause of traffic accidents,which poses a great threat to public safety and property.In order to reduce the losses caused by fatigue driving,many researchers have devoted themselves to the study about fatigue driving,such as driver behavior monitoring,brainwave monitoring,eye tracking and facial expression analysis.Each of these methods has its own advantages and disadvantages.Behavioral monitoring reflects the fatigue state by analyzing the driver's driving habits and facial expression,which is easy to operate but prone to be affected by the external environment.Brainwave monitoring is more accurate and can detect the fatigue state in real time,but the equipment is complicated and costly,which limits its large-scale application.The detection based on eye-tracking and facial expression analysis also has a certain potential for application,but errors may occur under different light conditions.Herein the review introduces the effects of fatigue on driving ability and compares the research results of various fatigue driving detection methods by searching,collating,analyzing and summarizing the relevant literatures at home and abroad.Various detection methods are analyzed and summarized,and it is pointed out that the fatigue driving detection method based on multi-feature information fusion will become a research hotspot.

An epilepsy prediction model based on two-dimensional convolutional neural network and Cox-Stuart test for non-independent patients is proposed to address the problem of how to effectively predict whether epilepsy patients are going to have an attack or not. After EEG data normalization and EEG signal noise removal using a notch filter and a high-pass filter,the filtered signals are inputted into the two-dimensional convolutional neural network model for feature extraction and classification,and Cox-Stuart test is used to determine whether an early stopping is needed or not,thereby reducing the computational and time complexities of the model. The model is tested under the conditions with pre-seizure periods of 10,30 and 60 min,respectively,and the results show that the model performs best when the pre-seizure period is 10 min. The model has an average accuracy,sensitivity and specificity of 97.70％,97.36％and 98.04％on the test set,demonstrating its superior performance.

Alzheimer's disease(AD)is a progressive,irreversible neurodegenerative disorder characterized by gradual brain cell degeneration,leading to progressive decline in cognitive function and ultimately death.Early identification and intervention are critical to AD diagnosis.In recent years,deep learning has further advanced image-based AD classification methods and facilitated the application of deep models in the early AD diagnosis.To achieve accurate early diagnosis and subsequent classification of AD,researchers have integrated deep learning with magnetic resonance imaging to develop more precise models.By analyzing and synthesizing relevant domestic and international literature,this review introduces commonly used public datasets and evaluation criteria for AD,analyzes the application of magnetic resonance imaging in AD classification and its integration with deep learning methods,and highlights the roles of techniques such as convolutional neural networks,transfer learning,attention mechanisms,and multimodal approaches in AD classification.It also discusses the advantages,limitations,and developmental trends of deep learning in AD classification,aiming to provide new insights for the application of deep learning in AD research.

The study aims to address the issues of class imbalance in sleep EEG data and gradient vanishing or explosion phenomena that may occur when deep networks extract more features.An improved adaptive synthetic sampling technique is firstly employed to perform data augmentation on the minority classes of sleep EEG data.Subsequently,convolutional neural networks and residual networks are utilized to learn data features,while a 3-layer bidirectional gated recurrent network is applied to explore deep temporal information and establish correlations between different sleep stages,enabling automatic feature learning and sleep cycle extraction.Finally,a multi-head self-attention mechanism is adopted to enhance the model's focus on critical parts of the sequence,thereby completing the classification of various sleep stages.Experimental results show that according to the AASM sleep staging criteria,the automatic sleep staging model integrating CNN-BiGRU and multi-head self attention achieves an overall accuracy of 90.77%and a Kappa coefficient of 0.88 on the Sleep-EDF-20 dataset after data class balancing,with the precision of N1 stage reaching 87.1%.On the Sleep-EDFx dataset,the model attains an MF1 score of 0.84 while maintaining a precision of 77.2%for N1 stage classification.These metrics demonstrate significant improvements in performance as compared with CNN-BiGRU model tested on the original dataset.When benchmarked against other related studies,the proposed architecture exhibits superior sleep stage classification accuracy.These findings collectively validate the effectiveness and generalization capability of the proposed method.

An epilepsy prediction model based on two-dimensional convolutional neural network and Cox-Stuart test for non-independent patients is proposed to address the problem of how to effectively predict whether epilepsy patients are going to have an attack or not. After EEG data normalization and EEG signal noise removal using a notch filter and a high-pass filter,the filtered signals are inputted into the two-dimensional convolutional neural network model for feature extraction and classification,and Cox-Stuart test is used to determine whether an early stopping is needed or not,thereby reducing the computational and time complexities of the model. The model is tested under the conditions with pre-seizure periods of 10,30 and 60 min,respectively,and the results show that the model performs best when the pre-seizure period is 10 min. The model has an average accuracy,sensitivity and specificity of 97.70％,97.36％and 98.04％on the test set,demonstrating its superior performance.

Alzheimer's disease(AD)is a progressive,irreversible neurodegenerative disorder characterized by gradual brain cell degeneration,leading to progressive decline in cognitive function and ultimately death.Early identification and intervention are critical to AD diagnosis.In recent years,deep learning has further advanced image-based AD classification methods and facilitated the application of deep models in the early AD diagnosis.To achieve accurate early diagnosis and subsequent classification of AD,researchers have integrated deep learning with magnetic resonance imaging to develop more precise models.By analyzing and synthesizing relevant domestic and international literature,this review introduces commonly used public datasets and evaluation criteria for AD,analyzes the application of magnetic resonance imaging in AD classification and its integration with deep learning methods,and highlights the roles of techniques such as convolutional neural networks,transfer learning,attention mechanisms,and multimodal approaches in AD classification.It also discusses the advantages,limitations,and developmental trends of deep learning in AD classification,aiming to provide new insights for the application of deep learning in AD research.

Fatigue driving is a major cause of traffic accidents,which poses a great threat to public safety and property.In order to reduce the losses caused by fatigue driving,many researchers have devoted themselves to the study about fatigue driving,such as driver behavior monitoring,brainwave monitoring,eye tracking and facial expression analysis.Each of these methods has its own advantages and disadvantages.Behavioral monitoring reflects the fatigue state by analyzing the driver's driving habits and facial expression,which is easy to operate but prone to be affected by the external environment.Brainwave monitoring is more accurate and can detect the fatigue state in real time,but the equipment is complicated and costly,which limits its large-scale application.The detection based on eye-tracking and facial expression analysis also has a certain potential for application,but errors may occur under different light conditions.Herein the review introduces the effects of fatigue on driving ability and compares the research results of various fatigue driving detection methods by searching,collating,analyzing and summarizing the relevant literatures at home and abroad.Various detection methods are analyzed and summarized,and it is pointed out that the fatigue driving detection method based on multi-feature information fusion will become a research hotspot.

The study aims to address the issues of class imbalance in sleep EEG data and gradient vanishing or explosion phenomena that may occur when deep networks extract more features.An improved adaptive synthetic sampling technique is firstly employed to perform data augmentation on the minority classes of sleep EEG data.Subsequently,convolutional neural networks and residual networks are utilized to learn data features,while a 3-layer bidirectional gated recurrent network is applied to explore deep temporal information and establish correlations between different sleep stages,enabling automatic feature learning and sleep cycle extraction.Finally,a multi-head self-attention mechanism is adopted to enhance the model's focus on critical parts of the sequence,thereby completing the classification of various sleep stages.Experimental results show that according to the AASM sleep staging criteria,the automatic sleep staging model integrating CNN-BiGRU and multi-head self attention achieves an overall accuracy of 90.77%and a Kappa coefficient of 0.88 on the Sleep-EDF-20 dataset after data class balancing,with the precision of N1 stage reaching 87.1%.On the Sleep-EDFx dataset,the model attains an MF1 score of 0.84 while maintaining a precision of 77.2%for N1 stage classification.These metrics demonstrate significant improvements in performance as compared with CNN-BiGRU model tested on the original dataset.When benchmarked against other related studies,the proposed architecture exhibits superior sleep stage classification accuracy.These findings collectively validate the effectiveness and generalization capability of the proposed method.

Objective: To investigate the characteristics and trends of malaria cases in Lishui City, Zhejiang Province from 2012 to 2023, so as to provide a basis for improving malaria prevention and control measures. Methods: Case data of malaria in Lishui City from 2012 to 2023 were collected from the Parasitic Disease Control Information Management System of the National Information System for Disease Control and Prevention in China. The parasite species, source of infection, temporal distribution, population distribution, geographical distribution, and clinical diagnosis and treatment of the cases were descriptively analyzed. Results: A total of 169 malaria cases were reported in Lishui City from 2012 to 2023, and P. falciparum malaria was the main type, accounting for 79.88% (135 cases). The positive rate of Plasmodium detection was 3.30‰(169/51 212), the highest was 5.41‰ (18/3 327) in 2017, and the lowest was 0.38‰ (1/2 632) in 2021. Malaria cases were reported in every month from 2012 to 2023, with 91 cases (53.85%) reported from May to October. There were 168 imported cases, of which 163 (96.45%) originated from Africa. There were 127 male cases (75.15%), and the majority of cases were aged 20 to 49 years, with 138 cases accounting for 81.65%. The majority of the occupation was overseas labor export workers, with 164 cases accounting for 97.04%. A total of 161 cases (95.27%) were registered residents of Lishui City, and cases were reported from all nine counties (cities, districts), with Qingtian County and Liandu District having the higher numbers of 98 and 41 cases, respectively. The median interval from onset to hospital visit for malaria cases was 2.00 (interquartile range, 4.00) days, and the median interval from hospital visit to diagnosis was 0 (interquartile range, 1.00) day. The diagnostic rate of first-diagnosed malaria cases in municipal and county medical institutions was 95.90% (117/122) and 91.49% (43/47), respectively, with no statistical significance (P>0.05). Conclusions The P. falciparum malaria was the predominant type in Lishui City from 2012 to 2023, with the majority of cases being imported. Male overseas labor export personnel aged 20 to <50 were the key demographic.

Objective To establish a hybrid deep learning lung sound classification model based on convolutional neural network(CNN)-long short-term memory(LSTM)for electronic auscultation.Methods Wavelet transform was used to extract features from the dataset,transforming lung sound signals into energy entropy,peak value and other features.On this basis,a classification model based on hybrid algorithm incorporating CNN and LSTM neural network was constructed.The features extracted by wavelet transform were input into CNN module to obtain the spatial features of the data,and then the temporal features were detected through LSTM module.The fusion of the two types of features enabled the classification of lung sounds through the model,thereby assisting in the diagnosis of pulmonary diseases.Results The accuracy rate and F1 score of CNN-LSTM hybrid model were significantly higher than those of other single models,reaching 0.948 and 0.950.Conclusion The proposed CNN-LSTM hybrid model demonstrates higher accuracy and more precise classification,showcasing broad potential application value in intelligent auscultation.