This PDF file contains the front matter associated with SPIE Proceedings Volume 13270, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

To alleviate the workload associated with image segmentation and enhance segmentation performance, there are still many details to be improved in the specific application of image segmentation. In particular, the Pulse-Coupled Neural Network (PCNN) presents challenges, such as determining the optimal selection of connection coefficients. Then an enhanced model known as the Memristor Pulse-Coupled Neural Network (M-PCNN) is proposed. This model integrates the connection coefficients from the traditional PCNN with resistance selection mechanisms. Firstly, the two memristor are operated in parallel to enhance the stability and fault tolerance of the model. If one memristor fails, the other can continue to work normally to ensure the stable operation of the system. Subsequently, the parallel configuration directly replaces the connection strength in the traditional model, and the interaction of two parallel memristors can extract the lesion area more effectively, making the segmentation result more accurate. Finally, the lesion region of the image was magnified and analyzed using both models. Experimental results demonstrate that, compared to the traditional PCNN model, the connection coefficients in the M-PCNN model are simpler to select, significantly reducing the workload associated with the segmentation task. Additionally, the M-PCNN model is more accurate in extracting targets from images. For some complex structures, M-PCNN model can better extract lesions. In this paper, four objective evaluation metrics are selected. Through the precise comparison of data, it is demonstrated that all M-PCNN models take better values to further improve the accuracy of medical diagnosis.

The model of transformer structure occupies a dominant position in the field of multimodal large model. While previous studies have highlighted the potential of Visual Transformer (ViT) models, their reliance on large datasets poses challenges in domains like medicine, where obtaining extensive data can be difficult. In such scenarios, traditional convolutional neural networks (CNNs) often outperform transformer-based models due to their ability to capture pixel level fine-grained information. In this paper, we proposed soft mask operation and fine-grained information awared visual transformer Med-T, a CNN-Transformer hybrid visual backbone network tailored for visual feature extraction task on limited datasets. Through extensive evaluation across three small datasets, Med-T consistently out performs alternative approaches, showcasing the efficacy of leveraging the pixel-level position information extraction ability of CNN branch.

Polyp segmentation has consistently been a difficult task because of the varying sizes of polyps and the significant intrinsic similarity between polyps and the surrounding tissues. To address the above problems, a contextual feature aggregation polyp segmentation algorithm combining Pyramid Vision Transformer and convolution (CFA-PVT) is proposed. Firstly, the Pyramid Vision Transformer is used to extract image global features, and the stage bridging module(SBM) is employed to enhance the ability of the network to handle polyp details and aggregate high-level polyp features. Subsequently, a feature enhancement module (FEM) is used to explore shallow polyp information. Finally, cross-layer feature fusion is performed by a global adaptive module (GAM) to realize feature interaction. This algorithm is evaluated on the CVC-ClinicDB and Kvasir-SEG datasets and further tested for generalization capability on the CVC-ColonDB dataset. The results demonstrate that the proposed method effectively segments colorectal polyp images, offering a new approach to diagnosing colorectal polyps.

In computed tomography (CT) systems, the presence of metal artifacts can significantly impair image quality, making subsequent diagnosis and treatment more challenging. Against the Metal Artifact Reduction(MAR) task, existing deep-learning methods have achieved satisfactory reconstruction results. However, most of the methods involve limited artifact-free prior knowledge to guarantee the fidelity of the final result. Additionally, the sinogram-domain information that many methods rely on introduces secondary noise, exacerbating the negative effects. To address these issues, we propose a novel Discretized Clinical Convergence Generative Network (DCCGN) that relies on image domain information and robustly introduces the clinical prior information in a quantitative manner to convert noisy features into clean ones to complete the reconstruction. Extensive experiments and evaluations have shown that DCCGN has superior generation and fidelity compared to several SOTA algorithms for both synthetic and clinical datasets.

The paravertebral muscle is a vital structure that maintains the stability of the lumbar spine. The decrease in muscle mass and the increase in fat content of the paravertebral muscle are closely related to the occurrence of various lumbar diseases. The degeneration of the paravertebral muscles is associated with various diseases, such as low back pain, degenerative scoliosis, osteoporosis, and so forth. At present, although a large number of studies have reported on vertebral computed tomography (CT) image segmentation, studies on paravertebral muscle segmentation are few. This study aimed to achieve the segmentation of the muscle region in vertebral CT images to provide clinically feasible index observation data for the diagnosis, treatment, and prognosis of related diseases. The traditional segmentation method is time-consuming and laborious. Also, the segmentation results vary significantly due to the different levels of experience of clinicians. Further, the repeatability is poor. This study used the automatic image segmentation model based on a deep learning algorithm, namely the U-net network model, to achieve vertebral muscle segmentation in CT images. The average Dice coefficient reached 0.9178, indicating a good segmentation effect of the segmentation model based on the U-net network. Based on the results of paravertebral muscle segmentation, automatic measurement and analysis of the cross-sectional area, paravertebral muscle density, and degree of fat infiltration can be further realized, there by guaranteeing the prognosis of patients with spinal diseases.

CT plays an irreplaceable role in modern medical diagnosis, and their quality directly affects doctors' judgment of diseases. Therefore, realizing the quality control of CT images to ensure that the image quality meets the diagnostic standards is an important measure to guarantee the medical quality and patient safety. In this paper, MATLAB is used as a tool to realize the automated detection of CT quality control, and the detection items include CT number of water, noise and layer thickness. The program is written to replace the manual detection steps to realize the measurement automation, and the interface is designed to form an APP through APP Designer. Experiments show that the automatic detection results of this software are more accurate and less time-consuming than the traditional way, which can be used for CT quality control detection.

CT is an essential tool in modern medical diagnostics and plays a pivotal role in the healthcare industry. Of which, the detection method has a great impact on the imaging effect and diagnostic effect. Regular calibration and verification of detection equipment can ensure stable performance, reduce errors, and ensure clarity and accuracy of imaging. At present, manual measurement is the primary method used in most regions of China. This approach not only results in laborious and inefficient work but also makes it vulnerable to subjective interference from staff members. It does not support effective quality control for CT scans and presents considerable obstacles for evaluating the quality of medical equipment. Based on these problems, this paper uses MATLAB to design and implement automated measurement of CT images.

Early prenatal ultrasound screening can significantly reduce neonatal mortality due to congenital heart disease(CHD). Due to the uniqueness of the fetal heart structures and the variety of fetal cases, the prenatal detection rate of fetal CHD is still quite low; therefore, an improved DenseNet is proposed to diagnose fetal CHD. Compared to the adult heart, the size of the fetal heart varies significantly and its location is not fixed, so a multi-scale feature fusion module is introduced into the network, which extracts multi-scale features in the fetal heart by combining convolutional kernels of various sizes. Secondly, there are complex structures and rich information in fetal ultrasound images, therefore the Efficient Channel Attention (ECA) mechanism is integrated into the network, which suppresses the expression of unimportant information and mentions the reliability of model classification. The experimental results demonstrate that the improved DenseNet achieves better results in the task of fetal CHD classification. Additionally, the improved DenseNet enhances the prenatal detection rate of fetal CHD by achieving the recall of 85% and the precision of 85.3% on the test set.

Esophageal cancer is the sixth leading cause of cancer-related mortality worldwide. Barium esophagram is an inexpensive, noninvasive, and widely available procedure. However, the variation in tumor size, the interference of surrounding tissues, and esophageal deformation make tumor detection very challenging. This paper proposes a novel algorithm for detecting esophageal cancer, consisting of a detection network and a classification network. A deformable residual bottleneck block is proposed to replace the residual bottleneck block in ResNet50 to sample the deformed esophagus adaptively. The improved ResNet50 is used as the backbone network of the detection network, which is combined with FPN network to predict tumors of various sizes. Then, regions of interest detected from each image of the barium esophagram by the detection network are used as the input of the classification network. The attention module CBAM is introduced into the classification network to enhance the significance of the esophageal region, and improve the classification accuracy. The algorithm is evaluated on 40 positive (1166 images) and 53 negative cases (1547 images), the accuracy of the algorithm on positive and negative cases are 100% (40/40) and 84.90% (45/53) respectively. The experiment demonstrates that our proposed method yields promising results with the barium esophagram dataset.

We introduced a novel multimodal radiomic feature fusion method based on game theory, specifically designed for glioma grading using multimodal magnetic resonance images. In a retrospective analysis, 257 patients (204 with high-grade gliomas [HGG] and 53 with low-grade gliomas [LGG]) were used as a training cohort, while internal and external test cohorts comprised 111 patients (88 HGG, 23 LGG) and 136 patients (114 HGG, 22 LGG), respectively. Imaging included T1-weighted, T2-weighted, fluid-attenuated inversion recovery, and contrast-enhanced T1-weighted sequences performed on 1.0 T, 1.5 T, and 3.0 T MR systems from public datasets and an additional 3.0 T MR system. Radiomic features were extracted from regions of interest across modalities. The proposed method leverages game theory to fuse multimodal MRI features, followed by a three-step feature selection process. Predictive models were subsequently built and validated on both internal and external cohorts. Area under the receiver operating characteristic curve (AUC), accuracy, F1 score, sensitivity, and specificity were calculated for evaluating the model performance. The fused feature model achieved an AUC of 0.953 (95% confidence interval: 0.906-1.000) for the internal test set and 0.853 (95% confidence interval: 0.777- 0.929) for the external test set. This feature fusion method shows promising robustness for glioma grading in radiomics using multimodal magnetic resonance images.

Purpose: The present study aimed to construct classification models for pulmonary adenocarcinoma using computed tomography (CT)‐based radiomics features and random forest method. Methods: A total of 289 patients with 295 lung adenocarcinomas were included in this study. A total of 1066 CT images were extracted. The final data set was randomized into the training set and validation set at the ratio of 80%:20%. A total of 1082 features were captured from a semi‐automatic segmentation method segmented lesion of a CT image. 9 optimal radiomic features obtained from root mean squared error (REMS) through cross validation and 14 radiographic characteristic features were selected to construct a random forest classification model. At the same time, compared with the results of the Support Vector Machine (SVM), Logistic Regression and C5.0 algorithm. Results: The area under the curve (AUC) scores of training feature set, radiographic characteristics feature set, and the optimal radiomic feature set for testing dataset were 0.974, 0.483, and 0.835, respectively, and the corresponding AUC values for validation dataset were 0.964, 0.915, and 0.841, separately. Conclusion: The developed random forest‐based classification models using radiomics features and radiographic features of CT showed a relatively acceptable performance in lung adenocarcinoma and could assist clinical rapid diagnosis and triage.

Introduction: Respiration-induced motion can lead to artifacts in chemical exchange saturation transfer (CEST) images, and inaccurate quantization of CEST signals. The aim of this study was to develop a motion-insensitive renal 4D (extra motion dimension) CEST imaging technique for clinical use during free breathing. Methods: We utilized a 3D turbo filed echo (TFE) sequence with in-plane golden-angle radial sampling and out-plane Cartesian sampling (stack of stars) to acquire renal CEST images from three normal volunteers under free breathing. The respiration states were separated, and XD-GRASP was used to reconstruct the CEST images. The efficacy of this method in suppressing motion artifact was evaluated. Results: The renal Z-spectrum demonstrated distinct CEST peaks, and no motion artifacts were observed in the CEST images. This motion-insensitive sequence proved to be reliable for renal CEST imaging. Conclusion: 4D CEST-MRI provides a motion in-sensitive CEST imaging approach for renal imaging in clinical nephropathy patients under free breathing, this may improve the accuracy of nephropathy detection.

It is of great clinical significance to distinguish the types of myopia into early prevention of vision deterioration in adolescents. Fundus tessellation (FT) in color fundus photography (CFP) is an important clinical characterization of early pathological lesions, which may indicate irreversible development of visual impairment. Therefore, early automatic detection and quantitative analysis of FT from CFP can help predict the vision disease progression and prognosis. However, the extremely complex morphology of FT images makes manual annotation extremely difficult and labor intensive. Therefore, it is necessary to study semi-supervised machine learning methods that do not require too much manual annotation to achieve automatic detection of FT. In this paper, we proposed a new semi-supervised end-to-end framework for tessellated fundus detection and segmentation of adolescents’ CFP images. This method incorporates co-decision models that combine different architectures to achieve accurate and automated FT detection results. Furthermore, we utilized an adaptive integration block (AIB) to do the best combination weights of different architectures and get a final prediction result. Finally, we applied our proposed framework to the clinical collected adolescents dataset for independent validation. By evaluating the parameters and comparing the segmentation visual results, it can be concluded that our proposed method can perform semi-supervised classification based on a small amount of labeled data, achieving the goal of effectively segmenting FT shadows.

Automated brain lesion segmentation using deep learning (DL) is crucial for understanding neurological diseases and aiding in diagnosis. Efficiently training a DL segmentation model often involves transfer learning (TL), where knowledge from a source task helps in the target task. Traditionally, this involves model weight transfer, initializing the target model with weights from a pretrained source model and then fine-tuning it with target data. However, this approach limits direct interaction between source data and the target task. This work introduces a new TL paradigm for brain lesion segmentation called Brain Lesion Transfer (BLeT). Instead of transferring model weights, BLeT directly utilizes source training data by transferring useful information to the target task. For a given target brain lesion segmentation task, it is assumed that annotated data for a similar source task is available. BLeT transfers lesions from the source training data to the target annotated data, creating additional, diverse training images, thereby enhancing the training of the target segmentation model. To address the challenge of different appearances in images from different tasks, BLeT includes a lesion appearance transformation method. This method adjusts the lesions from the source task to be compatible with the target images. Experiments on public datasets demonstrate that BLeT outperforms conventional TL methods based on model weights for brain lesion segmentation.

The research employs EEG in combination with the 1DCNN-LSTM algorithm for the purposes of identifying and classifying people with varying risks of schizophrenia. This collection of data for schizophrenia's high-risk and low-risk groups comprises EEG signals from 98 people, the division was made between a group of 50 with normal EEG signals and another group of 48 with high-risk indicators for schizophrenia. The duration of measurement for every participant stood at 12 minutes. The data set was processed in advance, encompassing tasks like channel exclusion, re-referencing, dividing the dataset, performing independent component analysis (ICA), windowing, normalizing, and segregating it into the testing, training, and validation sets. Subsequently, the processed EEG data was integrated into the 1DCNN-LSTMclassification model, where post-extensive learning, the model's weights were derived. The categorization system attained a 94.26% precision rate in the identification of the complete-channel EEG data collection. This research utilized the 1DCNN-LSTM algorithm to classify and identify high-risk and low-risk groups for schizophrenia, showcasing an adequate recognition capability that satisfies real-world application criteria. The system precisely categorizes populations at varying risk levels for schizophrenia through comprehensive EEG data, thus facilitating precise schizophrenia detection and offering prompt diagnosis and treatment for those at high risk for the condition.

Resting EEG microstates reflect the spatiotemporal dynamics features of brain activity, correspond to different restingstate functional networks, and are associated with cognitive functions. This study investigated the relationship between eyes-closed resting-state EEG microstates and cognitive functions in healthy adults, using the LEMON dataset with 168 participants. Through T-AAHC clustering, we found that five microstate classes best represent adult resting-state EEG. Correlation and regression analyses, along with structural equation modeling, indicated significant associations between certain microstate metrics and cognitive functions. Notably, the mean duration of microstate A was positively correlated with cognitive flexibility and fluid intelligence, while its occurrence and coverage were also linked to cognitive flexibility. The occurrence of microstate B was positively correlated with crystallized intelligence. Specific transition probabilities between microstates C, D, and E are negatively correlated with crystallized intelligence and fluid intelligence. In addition, all regression results were significant, and the structural equation models A and B were valid. Therefore, we propose that resting-state EEG microstates can predict individual cognitive functions.

Acupuncture and moxibustion act on acupoint areas with different frequencies, evoking a large number of responding activity of neurons to achieve the purpose of regulating human body functions. In the process of acupuncture, different frequencies of acupuncture evoked different neuronal spiking activity. In order to study the mechanism of acupuncture with different frequencies, Bayesian statistical model is used to optimize the results of the traditional classification algorithm based on spiking waveforms, which greatly reduces the missed detection rate of acupuncture responding activity. Then, the spiking events evoked by acupuncture at different frequencies were statistically analyzed, and the results showed that the number of neuronal spikes gradually increased with the increase of frequency. However, when the stimulation was increased to 120 times/min, the increase in the stimulation frequency will not evoke more spikes due to the saturation of frequency adaptation of the neurons. Finally, a probabilistic statistical model was used to encode the neuronal responding activity evoked by different acupuncture, and the maximum likelihood estimation method was used to fit the model parameters. The results show that the coupling parameters of stimulus are significantly smaller than the coupling parameters of spike-history, and the more the historical spikes, the smaller the coupling parameters of stimulus. This suggests that since acupuncture is a low-frequency mechanical stimulation, a large number of historical spikes in the spiking activity are the main factors that evoke the neuronal response. Thus, revealing the responding mechanism of different acupuncture frequencies.

Brain-computer interface technology (BCI) enables users to directly control external devices by establishing an information transmission path between the brain and external devices. Brain-computer interfaces based on the motor imagination paradigm have also begun to enter various fields. Therefore, the research on the brain-computer interface encoding and decoding algorithm of the motor imagination paradigm is particularly important. This paper proposes a model based on attention mechanism CBAM and EEGNet to classify motor imagination electroencephalogram signals (MI-EEG), and verified it on a public data set. Compared with a single EEGNet model, it improved by 3.7%, which is 8.1% higher than the traditional FBCSP model. The experimental results show the effectiveness of the new CBAM-EEGNet model on the four classification tasks of motor imagery.

Attention is closely related to human life. To detect attention states quickly and accurately with fewer resources, this research proposes a method for attention state detection, it is based on differential entropy (DE) and power spectral density (PSD). Electroencephalogram (EEG) data is from 15 participants. It was processed using the Fast Fourier Transform (FFT) to extract DE and PSD features, which was normalized. These features were input into a Support Vector Machine (SVM). After optimizing the model parameters, it achieved a well-performing attention state detection model. The proposed method achieved a maximum classification accuracy of 85% and an average accuracy of 67%, The model described in the statement surpasses traditional SVM models that are trained solely on DE or PSD features, as well as single-channel or multi-channel SVM models. The new method can be used to learn additional features for attention verification and generalizes well for the task of developing a robust deep learning system.

Sleep respiratory signals are among the crucial physiological parameters of the human body, playing a significant role in the diagnosis and treatment of sleep-related disorders such as sleep apnea syndrome. However, current detection methods have certain limitations and are not yet widely applied in practice. This paper presents a comparative experimental study of contact-based piezoelectric and non-contact millimeter-wave radar methods for sleep respiratory detection. The aim is to analyze the differences and characteristics of these methods in clinical applications, providing valuable references for the detection of sleep respiratory signals in clinical practice. Experimental outcomes indicate that the millimeter-wave radar acquisition technique manifests considerable disparities in the intensity of respiratory signals across distinct sleeping orientations, specifically ranking in the order of prone > supine > lateral. This observation underscores the method's heightened responsiveness to alterations in sleeping posture. Notably, when subjects are in the prone position, the peak values of respiratory signals acquired through this technique exhibit a higher degree of consistency. In contrast, the piezoelectric approach yields respiratory signal intensities that remain largely invariant across diverse sleeping postures, thereby affirming its robust stability amidst changes in posture. Additionally, it is noteworthy that the piezoelectric method facilitates the collection of respiratory signals with more stable peak values during both supine and lateral sleeping postures.

With the continuous development of the brain-computer interface (BCI) technology, the lower limb rehabilitation system based on Motor Imagery (MI) has gradually become a research hotspot in the field of rehabilitation. To recognize the lower limbs MI, this paper designed an experimental paradigm for MI lower limb MI and used the 1D-CNN-LSTM deep learning algorithm to classify lower limb movement features from MI EEG signals. Compared with classical machine learning algorithms, the results showed that 1D-CNN-LSTM has relatively higher accuracy. Meanwhile, the paper built a real-time lower limb rehabilitation system based on the 1D-CNN-LSTM algorithm, which verifies the effectiveness and feasibility of the algorithm. The system provides an advanced and effective solution for brain-computer interfaces based on MI.

The electroencephalogram (EEG) is vital for analyzing brain electrical activity in medical diagnosis and research. Aiming at the limitations of large size, high price, and inability to monitor EEG activity daily, a wireless wearable EEG measuring device was designed for patients with epilepsy. The designed system, with its low-power Bluetooth module, EEG acquisition module, and motion module, is not only efficient but also user-friendly. The system works at a sampling rate of 250 Hz for 8 channels and transmits data to a host computer or cell phone via Bluetooth. In addition, head movements are also recorded for behavior analysis. The results showed that the designed system has low noise and high-resolution performance, meeting the requirements for daily EEG measurement. The key benefit of this new device is its convenience and efficiency, providing a more user-friendly and effective tool for EEG monitoring, which could benefit seizure recording for patients with epilepsy in a home environment.

This study aims to identify the epileptogenic zone (EZ) during the interictal period in epilepsy patients using electrocorticography data from four individuals. The proposed localization method, which constructs two brain connectivity networks: autoregressive and directed transfer function networks, holds significant potential. Network features are extracted using graph theory techniques employed in machine learning models to classify electrode locations as either part of the EZ. Six node features from the directed graph are selected: indegree, outdegree, cluster coefficient, PageRank, hubs, and community. A balanced support vector machine (SVM) addressed data imbalance. The balanced SVM method achieves the accuracy, precision, and recall of 0.775, 0.475, and 0.554, respectively. The results suggest that the node features of the epileptic network may provide critical information for clinical EZ localization, offering a promising avenue for future research and clinical practice.

This study investigates the potential application of multimodal connectome techniques in predicting individual inhibitory control abilities. Utilizing comprehensive datasets from the UCLA Consortium for Neuropsychiatric Phenomics, which include both structural and functional connectivity data, this research aims to determine whether individual differences in inhibitory control cognitive functions are attributable to variations in these connectomes and whether the spatial distributions of different modalities of connectomes overlap. Inhibitory control abilities were measured using a computerized Stroop task, and whole-brain structural and functional connectomes were constructed by integrating resting-state functional magnetic resonance imaging (rs-fMRI) and diffusion tensor imaging (DTI). By employing Connectome-based Predictive Modeling (CPM) and leave-one-out cross-validation linear regression models, this study seeks to analyze the relationship between brain connectomes and inhibitory control performance. The results demonstrate that models based on either structural or functional connectomes can effectively predict individual inhibitory control abilities. Functional connectomes showed higher correlations in predicting positive networks, whereas structural connectomes exhibited stronger correlations in predicting negative networks. These findings highlight the critical role of brain structural and functional networks in supporting cognitive control and suggest distinct mechanisms of different networks in cognitive tasks. This study establishes the significant application value of multimodal connectome techniques in precisely predicting individual cognitive functions, providing an innovative research approach for the field of cognitive neuroscience. Although current findings require further validation and expansion, this work lays a solid foundation for utilizing connectome techniques to deeply understand and predict complex cognitive functions, opening new avenues for future clinical and scientific research.

Pneumonia is the leading cause of death from infection in children worldwide. Auscultation has unique advantages in the preliminary screening and follow-up examination of pneumonia due to its non-invasive and radiation-free nature. However, the accuracy of auscultation largely depends on the clinician's experience. Therefore, this study utilizes largescale surface respiratory sound signals for pneumonia screening and visualization to assist doctors in clinical diagnosis. A Butterworth filter is applied to the respiratory sound signals to reduce interference from heart sounds and environmental noise. Short-time Fourier transform (STFT) and Mel filter banks are used to obtain spectrograms and Mel-spectrograms of the respiratory sound signals. These spectrograms and Mel-spectrograms are fed into a convolutional neural network (CNN) to extract deep features and perform features fusion. A support vector machine (SVM) is used as a classifier for binary detection of the respiratory sound signals. Finally, the detection results are visualized in combination with multi-channel positional information. The computational results demonstrate that the predicted pneumonia lesion areas are consistent with the actual lesion areas, achieving the detection rate for pneumonia is 100%, the mean TTAR is 0.32, and the mean TPAR is 0.29. This study employs large-scale respiratory sound signals for the visualization of pneumonia lesion areas, enhancing the intuitiveness of diagnosis. It has significant advantage and potential in environments with limited medical resources and in follow-up examinations for pneumonia.

Objective: Depression is a serious psychiatric illness, which seriously affects the life and property safety of patients. However, the clinical diagnostic gold standard is still lacking, and the pathological mechanism is still unclear. This study used machine learning methods to conduct classification prediction research and attempt to explore the pathogenesis of depression. Methods: This study concluded resting-state functional magnetic resonance images from 51 patients with mild depression and 21 healthy control subjects in the OpenNeuro dataset. Multi-domain brain connectomic features, including FOurdimensional Consistency of local neural Activities (FOCA), functional connectivity density (FCD), and degree centrality (DC) were calculated and selected for support vector machine (SVM) model to distinguish mild depression from healthy controls. Results: After statistical analysis and feature selection, 10 brain regions were used as features for machine learning model construction. Most of these brain regions are located in the default mode network (DMN) network. The accuracy of the SVM classification model is 81.9%. Conclusion: The number of important features that are effective in identifying depression is less than 10. The SVM model with multi-domain brain connectomic features has good classification performance for distinguishing depression.

In the field of medicine, disease prevention is more important than treatment. Diabetes, as one of the diseases that are harmful and have a large number of patients, the prediction of diabetes using learning models is an essential part of diabetes prevention and treatment in the future medical field. In this study, compound feature selection was used to screen out the eight features with the most predictive ability, and diabetes was predicted by six machine learning models and two deep learning models, and the final results were obtained as follows: the XGBoost classification had the best prediction performance, with an accuracy of 99.1%, a precision of 99.0%, a recall rate of 99.2%, an F1 score of 99.1%, and an AUC value of up to 0.991; CatBoost and LightGBM models have the next best performance. This is consistent with our previous findings. This study further confirms the value and potential of the XGBoost model for diabetes prediction, identifying superior feature selection methods compared to previous studies, and improving the predictive performance of the model while reducing model complexity when dealing with more complex data.

Depression is one of the more prevalent mental health disorders, characterized primarily by low mood. Recently, there has been a striking shift, and increasingly younger demographics have started showing symptoms of the same, particularly in adolescents. This paper proposes a new method for the automatic detection of incipient adolescent depression by use of deep multimodal learning techniques. This aims at improving preparedness to better face the rising problem of adolescent depression. In the proposed approach, unimodal features are extracted from electroencephalography (EEG), electrocardiogram (ECG), and speech signals using the Transformer model and subsequently fused into a comprehensive multimodal feature set for binary classification. The model does not only increase its generalizability by fusing different physiological signals but also increases the accuracy and reliability of diagnostic results by fusing multimodal features.

Background: Cardiovascular disease is one of the leading causes of death worldwide. Electrocardiogram(ECG) signals play a crucial role in diagnosing various heart conditions, including arrhythmias and myocardial infarctions. There is a need for a reliable and efficient method to quickly identify abnormal heartbeats to aid early diagnosis and treatment. Methods: The study utilized the MIT-BIH arrhythmia database, which includes 48 groups of two-lead ECG signals. High-dimensional features were extracted from the ECG signals using the ts fresh package in Python. Feature selection was performed using variance analysis, Spearman correlation, mRMR, and LASSO methods. Logistic regression models were then constructed to predict abnormal heartbeats. Results: The final model included 10 key features and demonstrated high diagnostic performance. The AUC was 0.958in the training set and 0.947 in the test set, with specificities of 0.930 and 0.851, and sensitivities of 0.881 and0.892, respectively. The model outperformed traditional methods and deep learning models such as CNN and VGG in identifying abnormal beats. Conclusions: This study presents a robust and effective nomogram model for distinguishing abnormal ECG signals, highlighting its significant clinical application potential. Future research will focus on expanding sample sizes and incorporating additional methods for feature calculation to further enhance model generalizability

In the process of education and learning, the lack the process of education and learning, the lackers' cognitive load may lead to the lack of accurate assessment of learners' cognitive load may lead to the design of teaching content, the provision of learning resources, and the choice of teaching methods that do not match the cognitive characteristics of learners, thereby causing difficulties in learning and affecting learning outcomes. To better achieve teaching objectives, this article reveals the level of cognitive load faced by participants in the process of searching for specific texts or symbols through tracking and in-depth analysis of eye parameters such as pupil diameter, gaze point, and gaze duration. The results show that when individuals face complex tasks, their pupils dilate sharply, a phenomenon stemming from the demand for more cognitive resources by complex tasks, which in turn triggers adjustments in the autonomic nervous system. In addition, high-frequency scanning behavior was also observed, and this frequent scanning behavior is precisely the user's attempt to quickly capture and process a large amount of information in a limited time, reflecting the tense allocation and efficient utilization of cognitive resources. The effective combination of the two can better assess the cognitive load level of the subjects.

Objective: Propose a method for modeling knee joint prostheses and comprise the construction of multiple sets of prostheses with varying geometries to investigate their wear properties. Methods: Multiple sets of total knee prostheses were built based on natural knee CT images by curve fitting method, and the contact and wear properties of the prostheses were simulated using finite element analysis. Results: The results of the simulation demonstrated that, for the same material, the prosthesis with a larger radius of the sagittal anterolateral contour curve of the tibial component exhibited superior wear performance, and the posteriorly inclined prosthesis demonstrated superior wear performance compared to the horizontal prosthesis. Under the same structure, the wear performance of the UHMWPE material was found to be superior to that of the PEEK material, regardless of the prosthesis structure. Conclusion: Tibial components with a larger radius of sagittal plane curve and tibial components with posterior inclination characteristics have better wear performance. The results of the study have a reference for knee prosthesis design and wear performance assessment.

Stenotic coronary artery disease (i.e. partial blockage of the arteries feeding the heart muscle) produces disturbed flow downstream from the blockage, interacting with the artery wall and generating low-amplitude audio-frequency vibrations. Some of this energy reaches the chest surface where it is detectable as an acoustic signature, distinct from heart sounds. We have performed in-vitro experiments on a model chest filled with a soft-tissue mimicking gel, covered with a polyurethane “skin” fitted with a variety of sensors, and within which is mounted a latex “artery” containing 3-D printed stenoses of various geometries. With this set-up, we have proved the principle that signals associated with the presence of a stenosis can be detected at the skin surface and have now developed a device consisting of an array of sensors incorporated into a stick-on chest-patch. The sensors transmit the signals wirelessly to a data capture unit from which the characteristics associated with disturbed flow can be identified. The time-domain signals are filtered, transformed to the frequency domain and the area under sections of the resulting spectra serve as the dependent (predicted) variable in a multivariable regression model where the independent variables are flow rate, frequency, stenosis geometry and sensor position. This has shown that the presence of stenosis-associated disturbed blood flow be detected, and its position and severity can also be inferred. We have now developed an improved patch and a validation trial will be carried out, initially on healthy volunteers and subsequently on patients with chest pain undergoing simultaneous diagnostic angiography.

For the implementation of sonothrombolysis, the acoustic pressure inside the blood vessel should be revealed according to the ultrasound burst and biological tissue conditions. The objective of this study is to measure the magnitude of the acoustic pressure inside blood vessel exposed to high-intensity focused ultrasound (HIFU), and to reveal its changing characteristics according to the ultrasound parameters (power and frequency) and tissue configurations. The tissue mimicking phantom with HIFU exposure was modeled to simulate the acoustic pressure. The results showed that for a biological tissue system composed of skin, fat, muscle, and blood, the peak pressure at the focus with blood zone increased as the insonation frequency increased (0.5-2 MHz). Pressure attenuation with respect to blood vessel depth(10-30 mm) intensified according to increment of HIFU power and frequency. Greater attenuation was observed when the frequency surpassed 1.1 MHz, varying with skin (1-5 mm) and fat tissue (2-7 mm) thicknesses. The results suggest that at frequencies below 1.1 MHz, identical HIFU power can be utilized for different individuals and lesions, there by achieving similar outcomes in clinical treatment.

The development of conductive hydrogels with excellent adhesion, self-healing and stretchable properties has become a key issue to be solved. In this work, hydrogels were successfully prepared by photopolymerization using amphoteric methacryloyl ethylsulfobetaine (SBMA) and nonionic acrylamide (AM) as monomers, and sodium chloride (NaCl) was added to enhance electrical conductivity of hydrogels. The prepared hydrogel has good self-healing property (selfhealing time is about 3h) and mechanical properties (tensile strength can reach 0.018 MPa, and the maximum tensile strain is >1200%). Due to the presence of SBMA and NaCl in the hydrogel, electrostatic interaction and hydrogen bonding in the hydrogel network are very favorable for migration of ions, so the hydrogel also has good electrical conductivity (2.1 S/m). In addition, the hydrogel also has strong adhesion ability, in the experimental process of adhesion to the substrate materials (including metal, rubber, glass, plastic), the four substrate materials are able to achieve excellent adhesion effect. In addition, the hydrogel has good durability under small tensile strain conditions, and can accurately monitor the movement signals of each joint of the human body. Due to the presence of free ions in the internal polymer network structure, the conductive principle is very similar to that of human skin, so the hydrogel has great potential in the field of biomedical and wearable flexible sensors.

Combining nerve electrical stimulation-based motion induction and electromyogram (EMG)-based motion decoding enables active rehabilitation and can potentially improve fine motor function in stroke patients. However, current commercial electrical stimulators lack the ability to switch between different stimulation sites, resulting in typically a single type of motion that cannot meet the needs of training for multiple hand motions. In addition, the stimulation current causes the EMG signals to contain stimulation artifacts, which directly affects the accuracy of motion decoding. In this study, we developed a closed-loop hand function rehabilitation training system based on peripheral nerve electrical stimulation and synchronous motion decoding. It consists of a high-density EMG acquisition module with real-time stimulation artifact removal and a multi-channel electrical stimulation module with a switching matrix. Its performance was experimentally tested and the results showed that the system can function faultlessly on healthy subjects. With this rehabilitation system, stroke patients can achieve active rehabilitation training with the unaffected hand driving the affected side, which can potentially help to change the neuroplasticity of brain and promote the recovery of motor function.

Hypertension, as an important risk factor for cardiovascular diseases, threatens the lives of adults worldwide every year. Therefore, continuous blood pressure (BP) monitoring is necessary for the prevention and early diagnosis of hypertension. To achieve cuffless BP monitoring, an end-to-end BP waveforms prediction model was implemented using photoplethysmography (PPG) and deep learning in this paper. The proposed model successfully predicted three blood pressure values: systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (MAP) from PPG waveforms using U-Net, and their mean absolute errors (MAE) and standard deviations (STD) were in the range of 4.54 ±7.42 mmHg, 2.47 ±4.62 mmHg, 1.67 ±3.65 mmHg, respectively, compared to the reference BP values. This provides a possibility to realize non-invasive continuous blood pressure monitoring in daily life through portable wearable devices.

Objective: The effect of arabinogalactan on immune function in mice has been systematically studied to provide experimental and theoretical basis for clinical and food development and application. Method: Kunming mice were randomly divided into four groups, and a mouse immunosuppressive model was established using cyclophosphamide as a tool drug. Blood routine, immune organs, liver cell physiological sections, and antigen physiological indicators (AST, ALT, MDA, TNF) in the blood were measured- α And IL- β). In vitro lymphocyte proliferation and transformation were also evaluated. Result: We successfully established a cyclophosphamide-induced immunosuppressive mouse model. In the arabinogalactan-treated and the cyclophosphamide-treated groups, increases were observed in body weight, thymus and spleen weight, red blood cells, white blood cells, and lymphocytes. The results of liver tissue sectioning and physiological index analysis showed that the arabinogalactan group could positively regulate various indicators compared to the control group, with significant differences; The positive regulation of arabinogalactan and cyclophosphamide groups was significant compared to the cyclophosphamide group, and the difference was extremely significant; The single use of arabinogalactan has the best effect on various indicators in mice. Conclusion: Arabinogalactan has an antagonistic effect on immunosuppressants, improves various indicators of mouse immune function, and has the effect of enhancing mouse immune function.

This research investigated the potential pharmacological mechanism of Bazhen decoction in the treatment of depression utilizing network pharmacology and molecular docking techniques. The study selected 160 effective active ingredients and their 4775 related targets in Bazhen decoction through the Traditional Chinese Medicine Systems Pharmacology Database (TCMSP), and with further analysis identified 270 potential targets for Bazhen decoction's anti-depressive effect. Cytoscape software helped construct an active ingredient-target network and protein-protein interaction(PPI)network, revealing core targets and key biological processes. GO function pathway enrichment analysis showed that biological processes with respect to the decoction’s anti-depressive effect mainly involves response to lipopolysaccharides, aging, and positive regulation of the apoptotic process. KEGG pathway analysis demonstrated that the primary pathways affected by Bazhen decoction include neuroactive ligand-receptor interaction, cAMP signaling pathway, etc. Molecular docking analysis indicated that key compounds in ginsenosides, among others, have good binding ability with key targets like TP53, providing molecular evidence for the anti-depressive effect. In conclusion, this research confirms that Bazhen decoction exerts anti-depressive effects through multiple components, targets and pathways, providing a scientific basis for its application in the treatment of depression and theoretical support for subsequent basic research and clinical applications.

Recent studies suggest that Cassia tora L. may have beneficial effects on colorectal cancer. However, the mechanisms and targets through which Cassia tora exerts its effects on colorectal cancer remain inadequately understood. This study aims to elucidate the active components, targets, and mechanisms of Cassia tora in colorectal cancer using network pharmacology and bioinformatics approaches. The results reveal that Cassia tora contains 18 active components associated with 266 targets relevant to colorectal cancer. The top 15 core targets identified include TP53, SRC, PTGS2, ESR1, CYP19A1, HSP90AA1, PIK3CA, TNF, EGFR, HRAS, MMP9, CASP3, UGT2B7, HSP90AB1, and CYP2C19. Differential expression analysis, survival analysis, single-cell transcriptome analysis, and drug sensitivity analysis indicate that PTGS2, TP53, and CASP3 are promising targets for pharmacological treatment of colorectal cancer. Additionally, this study reveals that the mechanisms by which Cassia tora exerts its effects on colorectal cancer involve pathways associated with prostate cancer, lipid metabolism and atherosclerosis, and endocrine resistance. In conclusion, this study is the first to employ network pharmacology and bioinformatics to elucidate the active components, targets, and mechanisms of Cassia tora in the context of colorectal cancer, offering new insights for the development of innovative therapies for this disease.

Objective: The potential targets and mechanism of action of Perilla seeds in the treatment of hypertension was explored by using network pharmacology. Methods: The seed components and related indicators of Perilla frutescens were screened through TCMSP database with OB value ≥ 30% and DL value ≥ 0.18, and the names of target proteins were standardized with the help of the UniProt database. Targets related to hypertension were screened out by GeneCards and OMIM disease databases, and were de-duplicated and sorted out, and then the common targets of drugs and diseases were extracted with the help of Venn diagrams. Imported the relevant files in Cytoscape 3.9.0 software and generated a component-target-disease network diagram about Perilla seeds for the treatment of hypertension. Generated a PPI network diagram using the String database and Cytoscape 3.9.0 software to identify the core targets of Perilla seeds for the treatment of hypertension. Applied the DAVID database for the GO function and KEGG pathway enrichment analysis, and speculated the main signaling pathways interfering with hypertension. AutodockTool 4 software was used to molecularly dock the core targets and components screened. Conclusion: Perilla seeds may be used to treat hypertensive disorders through components such as lignans and β-sitosterol, which act on target genes such as AKT1, TNF and IL6, and regulate the PI3K/Akt pathway and AGE-RAGE pathway. The results showed that Perilla seeds can be used to treat hypertension in a variety of ways, such as regulating the PI3K/Akt pathway and the AGE-RAGE pathway.

Objective: Dual Heart Disease in Traditional Chinese Medicine (DHDTCM) is characterized by suboptimal heart function and depressive-like behaviors. Our hypothesis was that electro-acupuncture (EA) could enhance heart function and alleviate depression in this setting. Methods: Sixty SD rats were randomly divided into five groups: blank, model, EA, traditional Chinese medicine(TCM), and Western medicine(WM). The DHDTCM model was created via Chronic Unpredictable Mild Stress (CUMS) and isoproterenol injection. In the EA group, five points were alternately stimulated by EA. Different treatments were given for 21 days: sertraline in the WM group, an herbal formula in the CM group, and normal saline in the model group. Results: The model was successfully established. EA improved depressive-like behaviors in dual heart disease, as shown in tests. In the OFT, relevant indicators increased in the EA group. EA enhanced cardiac function, as indicated by changes in CD31 and the ECG ST segment. Levels of BDNF and NGF rose. EA regulated HIF-1α content in relevant parts. Conclusions: The results suggest that EA can enhance heart function and ease the depressive state in rats. It indicates that the therapeutic effect of EA on dual heart diseases might result from regulating HIF-1α.

Using density functional theory at the level of oniom (B3LYP/6-31g(d,p):UFF), we investigated the chiral transition process of the Methionine molecule in 1F2-MOL. Furthermore, a complete chiral transition path reaction potential energy surface was drawn by looking for the extreme value point structures including the transition state and intermediate. The results show that the hydrogen atom on the chiral carbon atom of S-Met@1F2-MOL- Methionine molecule can transfer to the other side of the carbon atom via the oxygen atoms of carboxyl atoms as abridge, to achieve the chiral transition of Methionine molecule from S-Met@1F2-MOL-type to R-Met@1F2-MOL-type.The biggest reaction energy barrier is 317.9480KJ/mol, derived from the third transition state TS1-R-Met@1F2-MOL; other corresponding products of the intermediate transition process will not be described systematically, due to space constraints, and the following work will continue. The reaction mechanism of the migration of 12H atoms between carboxyl 9O and chiral 1C is obtained through 2 intermediates and 3 transition states under the condition of limiting molecular sieve, which further improves the theoretical system of chiral transition mechanism of Met.

Implicit expression H2O solvent S-Phe molecular system(S-Phe-1) electron-hole diagram analysis is not reported, for this reason, the excited state wave function of S-Phe-electronic structure characteristics of S-Phe-1. Based on the PBE0 method, TDDFT electronic excitation calculation is made using 6-311G(d) and defTZVP three different base groups, and the electron-hole analysis and analysis of S-Phe-1 base state S0 to S1, S2,S3,S4,S5,S6,S7 excitation electron-hole analysis is given. It is show that: through electron and hole diagrams, the electron excitation characteristics of the S-Phe-1 molecular system are demonstrated, and the two base groups can see that the analysis of S1,S2,S3,S4,S5,S6 excitation electron excitation characteristics is qualitatively consistent, and the results of the analysis are basically the same;S7 excitation state is different, with polarization function of the 6-311G (d) base group and defTZVP base group is different, the former refers to the benzene-based to the niobium-based pi→pi* charge transfer excitation, the latter is the amino-topyridine n→pi* field excitation.

In recent years, progress in deep learning has significantly refined AD/MCI classification, but the relationship between functional connectivity changes and structural connectors remains to be created. To address this issue, this article proposes an inventive diagnostic system that utilizes the brain's effective connectivity network and integrates the Graph Attention Network (GAT) with the Long Short-Term Memory Network (LSTM). with the Long Short-Term Memory Organize (LSTM). By capturing brain interactions and dynamic changes, the framework can progress with demonstrative precision. Utilizing the Alzheimer's Malady Neuroimaging Activity (ADNI) dataset, the framework proved to be excellent at recognizing and predicting Alzheimer's disease, which illustrates the clinical potential it has. This research details the design, implementation and initial validation of the proposed method, emphasizing its effectiveness.

Reconstructing gene regulatory networks (GRNs) is a fundamental challenge in bioinformatics that aims to unravel the complex relationships between genes and their regulators. Graph convolutional neural networks have shown more significant improvements in this field than traditional methods. However, GCNs rely heavily on smooth node features rather than graph structures. To address this limitation, Two-layer Neighbor Overlapping Perceptual Graph Convolution Network (Tnop-GCN) is proposed, that jointly learns local and global structural features by PageRank and DeepWalk. Experiments on DREAM4 dataset demonstrate that Tnop-GCN outperforms many other gene regulatory network reconstruction methods.

In the field of drug addiction, current assessment of drug addiction degree still relies on behavioral scales. However, this method has some factors such as patient's deliberate concealment, operator's subjective judgment and lack of neurophysiological indicators. To overcome the shortcomings of traditional behavior assessment, this study constructs an EEG-based deep learning model, called MCFBNet (Multi-Convolutional Filter Bank Network). The MCFBNet not only addresses the issue of poor generalization in small sample sizes but also proposes a novel approach in the analysis of drug addiction severity. This study, with the goal of enhancing classification accuracy, employs feature augmentation and channel perturbation techniques to increase the model's temporal coverage, thereby overcoming the shortcomings of traditional deep learning networks in assessing drug addiction. Cross-validation was conducted on data from patients with Methamphetamine Use Disorder (MUD), and ablation analysis was performed on the number of filters and the temporal coverage of data dimensions to determine the optimal values. An optimal analysis was also conducted on the model parameter update method. The classification results of the MCFBNet were visualized using a confusion matrix, and the experimental results were compared with other deep learning models and traditional machine learning models, yielding competitive outcomes.

Objective: The effectiveness of an analysis system of a continuous glucose monitoring (CGM) in the management of Type 2 Diabetes Mellitus (T2DM), focusing on the quality of glycemic control via sophisticated time series analysis. Methods: Approximate entropy (ApEn) and other entropies were used for analysis in the post hoc analysis of 858 subjects with T2MD. Spearman's correlation coefficient was calculated between entropy values and selected physiologic indicators to verify the possible clinical validity of these indicators. Main results: The application of entropy analysis enhanced the quantification of glycemic control complexity. Correlation for glycated hemoglobin A1c (HbA1c), glycated albumin (GA) was demonstrated with ApEn: −0.40, and −0.39, while the correlation coefficient for sample entropy (SampEn) was −0.29, and −0.26, respectively (all P < 0.001). These large negative correlations confirmed the validity of entropy measures in interpreting CGM data. Correlation analysis between entropy measures and metrics such as HbA1c highlighted the potential of this approach to provide insights into diabetes management. Conclusions: The use of entropy analysis has theoretically enriched the methodology for analyzing CGM measurements and provided a valuable tool for clinical practice. The method improves the management of glycemic control among patients with diabetes, potentially influencing personalized treatment strategies and improving overall diabetes care.

Background: Pulsed electric field (PEF) ablation has recently been applied by researchers in the treatment of atrial fibrillation (AF), which is able to utilize high-voltage electric fields to produce damage to the myocardium for the purpose of treating AF. The effect of epicardial fat layer on the ablation effect has not been systematically studied. The purpose of our study was to establish a computer simulation model to rationally simplify the human organ, which was used to evaluate the effect of the fat layer on the ablation damage area. Methods: Firstly, by building a computational simulation model, different tissues of the heart were simplified and a three-dimensional computational model containing only the ablation device of interest was built, the ablation damage range was assessed in the post-processing interface using an electric field threshold of 400v/cm, and finally, the thickness of the fat layer was varied in order to assess the effect of the fat layer on the ablation area. Results: The epicardial fat layer had a weakening effect on PFA, and the total ablation depth decreased when the thickness of the fat layer increased, and when the fat layer reached 1.5 mm, the total ablation depth stabilized and the ablation depth of the myocardial layer decreased. In addition, we found a strong linear relationship between the total ablation depth and pulse voltage.