Abstract:Aiming at the problems of cumbersome process and high cost of existing early screening methods for Alzheimer's disease(AD),we proposed a cognitive impairment recognition model based on machine learning.Firstly,we extracted the acoustic and seman-tic features of speech signals,combined with the Shanghai cognitive screening(SCS)scale test scores and personal information to im-prove the accuracy of identifying cognitive impairments.Then,Shapley additive explanations(SHAP)and local interpretable model-ag-nostic explanations(LIME)were used to analyze the importance of different speech features to enhance the interpretability of the mod-el.The results of manual Q&A dataset showed that the recognition accuracy,area under the receiver operating characteristic curve(AUC)and F1 score for cognitive impairments reached 0.9000,0.9023 and 0.9000,respectively.The study can provide new insight for early detection of AD.

Abstract:To solve the problem of low accuracy when extracting respiratory rate(RR)from photoplethysmography(PPG)signals,we developed an embedded RR detection system.The system included pulse wave acquisition module,software control system,upper computer interface design,WeChat small program design and ensemble empirical mode decomposition combined with wavelet threshold(EEMD-WT)respiratory wave reconstruction algorithm design to realize real-time monitoring of patients' RR.Compared with the Min-dray ePM10 system,this system could obtain the PPG and reconstructed respiratory signals of the measured person under the condition of ensuring high accuracy and low power consumption,and accurately calculate the RR and display.The experimental results showed that the root mean squared error(RMSE)of RR reconstruction was 0.2147.This study can provide an effective solution for reconstruc-ting PPG respiratory waves.

Abstract:Hypertension is one of the most common cardiovascular diseases,continuous blood pressure monitoring is essential for early detection and management of hypertension.Photoplethysmography(PPG)as a noninvasive physiological sensing technique,has been widely used in medical monitoring.Recent studies show that integrating deep learning with PPG signals enables effective blood pressure estimation and improves real-time monitoring accuracy.This review systematically summarizes the latest research progress of deep learning models such as Transformers,graph neural network(GNN)and transfer learning,and compares the application charac-teristics and performance of each model in PPG modeling.Finally,the advantages and disadvantages of various models are summarized,and future research directions are proposed to provide references and inspirations for non-invasive prediction and intelligent monitoring of hypertension.

Abstract:To explore the potential application of poly(D,L-lactic acid)(PDLLA)in tissue repair,we used PDLLA as the sub-strate material and prepared the precursor solution by mixing chloroform and N,N-dimethylformamide(CF/DMF)as the solvent.PDL-LA absorbable barrier membranes were fabricated via the electrospinning process.The membrane exhibited a most probable pore size of 1.08 μm and porosity of 64.6%,which could provide appropriate channels for the transport of nutrients and excretion of metabolites.Mechanical test results showed that the membrane had a tensile strength of up to 7.09 MPa,elongation at break of 129%,and tearing force of 4.39 N,which could meet the mechanical requirements for tissue barrier membranes.In vitro degradation experiments con-firmed that the PDLLA membrane could maintain stable mechanical properties within a 3-month degradation period,ensuring the sus-tained and effective performance of the tissue isolation function.This study can provide a basis for the clinical application of PDLLA ab-sorbable isolation membranes.

Abstract:In order to continuously monitor and feedback the parameters of external chest compression during cardiopulmonary re-suscitation(CPR),we designed and manufactured a non-contact CPR quality monitoring device based on depth camera,which could obtain the depth information of the hand of the compressors based on depth camera and convolutional neural network without contact with the emergency patient.Additionly,through coordinate transformation and peak search algorithms,parameters such as compression depth and frequency were obtained.To verify the performance of the equipment,we built an evaluation system and measured the simu-lated chest compression datas.The results showed that the absolute average error of measuring the depth of compressions was 2.0 mum and the absolute average error of compressions frequency was 0.1 compressions per minute.The equipment has high detection accuracy and good robustness,which can meet the clinical needs.

Abstract:Aiming at the problem of insufficient use of hierarchical feature information in the feature extraction process of traditional convolutional neural networks due to scale limitations,we proposed a multi-label classification model XResT for electrocardiogram(ECG)signals based on deep feature fusion using residual convolutional network and Transformer encoder.Firstly,the XResT em-ployed an improved ResNet101 backbone network to achieve deep mining of multi-level local features of ECG signals.Subsequently,the Transformer encoder was adopted for global feature enhancement and a cross-layer coordinate attention was designed to achieve dy-namic fusion of local and global features.Comparative experiments on the CPSC-2018 dataset demonstrated that while the accuracy rate of this model was 96.43％,the average F1 score of multi-label classification reached 82.87％,a 3％improvement over existing bench-mark models.The ablation experiment verified the effectiveness of the cross-layer attention mechanism for complex heart rhythm char-acteristics such as ventricular premature beats.The feature fusion framework proposed in this study provides new insights for multi-level feature extraction of ECG signals,significantly improving the clinical applicability of automatic arrhythmia diagnosis systems.

Abstract:To investigate the effect of multi-point injection of Fe3O4 nanoparticles on the ablation results,we constructed a mathe-matical model of Fe3O4 nanoparticle-assisted microwave ablation(MWA)in liver tissue by COMSOL Multiphysics 6.1 software.First-ly,the effect of the number of injection domains(2～7)on the ablation zone of MWA was investigated when the injection volume was fixed,and after that,the effect of different injection radii(1～4.49 mm)on the ablation zone of MWA was investigated when the num-ber of injection domains was fixed.The results showed that when the injection volume was fixed at 2.45 GHz and 15 W,the number of injection fields was 6,7 to achieve better ablation effect.When the number of injection fields was 6,the injection radius was 3.62 mm,the ideal ablation effect was achieved.The transverse diameter of the ablation zone increased by 5.77 mm,the volume increased by 68.75％,and the roundness increased from 0.786 to 0.885.When the number of injection fields was 7,the radius of the injection field was 3.49 mm,the ablation effect was better,the transverse diameter of the ablation zone was increased by 3.52 mm,the volume was increased by 61.98％,and the roundness was increased from 0.786 to 0.869.The research indicates that adding Fe3O4 nanoparticles to MWA can increase the sphericity of the ablation zone,which is more suitable for the treatment of spheroidal tumors.

Abstract:To achieve early prediction of sepsis,we designed a prediction model KA-Transformer based on time series data.A ker-nel attention mechanism was introduced in the KA-Transformer to improve issues such as limited training samples,numerous parame-ters,and uneven sample distribution.With the input of continuous time series data,at three prediction time points 1,6,and 12 h before sepsis onset,the area under the receiver operating characteristic curve of the model for predicting sepsis were 0.962,0.944 and 0.984,respectively,the accuracy rates were 92.3％,93.9％and 96.1％,respectively.The experimental results show that the KA-Transformer significantly outperforms existing methods in terms of accuracy and generalization ability for sepsis prediction,and has the potential to enhance the timeliness and reliability of prediction for sepsis prediction.

Abstract:To improve the segmentation accuracy and efficiency of breast tumors under limited computing resources,we proposed a tumor segmentation algorithm for mammograms by improving the YOLOv8 model.Firstly,a feature extraction module PC-C3K2 was de-signed by pinwheel convolution(PConv)and C3K2 module to significantly expand the receptive field and reduce the number of model parameters.Secondly,the spatial channel decoupling downsampling module was utilized for downsampling,and the computational com-plexity was reduced by independently processing the spatial information and channel information.The experimental results on the IN-breast and the CBIS-DDSM datasets showed that the mAP50 value of the proposed algorithm reached 91.8%,and the number of param-eters could be reduced by 28%.The algorithm can significantly reduce the number of model parameters while improving the tumor seg-mentation accuracy of mammograms,which is of great significance for the early screening and diagnosis of breast tumors.

Abstract:Tendon injury is one of the common sports injury disorders.Traditional methods for treating tendon injury are prone to problems such as tissue adhesion and scar formation.Electrospinning technology can prepare nanofiber membranes with extracellular matrix(ECM)biomimetic fibrous structure and various drug encapsulation,which can provide a suitable biological microenvironment for tendon cell growth,reduce the risks of inflammation,infection and adhesion,and effectively restore the mechanical properties of in-jured tendons.Based on this,this paper analyzes and summarizes the application research and prospects of electrospun fiber stents in tendon injury repair covering material selection,structural design,and drug encapsulation to provide a reference for the repair and treatment of tendon injury.

Abstract:Aiming at the problems of poor real-time and interpretability of sepsis prediction model based on machine learning,we designed a sepsis real-time prediction model with high timeliness and clinical explainability.Among them,the real-time prediction module could quickly obtain the 3 h dynamic feature sequence of non-invasive physiological indicators,and calculated the mean,standard deviation and end value.The interpretation module introduced Shapley additive interpretation method(TreeSHAP)based on tree structure,which could comprehensively improve the interpretability of the real-time prediction model of sepsis from the perspective of single prediction and global interpretation.The result showed that the accuracy,sensitivity and area under the curve of the sepsis real-time prediction model reached 0.71(95%CI,0.69～0.73),0.71(95%CI,0.70～0.73)and 0.76(95%CI,0.75～0.77),respec-tively.This model can not only provide real-time dynamic early warning for sepsis in critically ill patients,but also help clinicians deeply understand the generated details and the overall logic of the model,improve the clinical credibility of the model,and offer sup-port for clinical decision-making.

Abstract:To address the issue that spatial information in 3D brain image data is complex and difficult to extract effective features,we proposed a 3D multimodal brain image reconstruction algorithm 3MbiRMA based on masked image modeling technology.A dual-branch structure was adopted to extract features of diffusion tensor imaging(DTI)and magnetic resonance imaging(MRI)3D brain image,to fuse multimodal features through feature decoupling,and to realize image reconstruction based on the decoder.Furthermore,the masking strategy and squeeze-space attention mechanism could not only significantly reduce information redundancy and enhance feature effectiveness,but also effectively reduce the computational complexity.Experimental results on the BeijingEN and ADNI data-sets indicated that the computational complexity of the 3MbiRMA was only 1/4 of that of the classical masked autoencoder(MAE)mod-el.Compared with the classical 3D reconstruction algorithms(3D UNet,VNet),the computational complexity of the 3MbiRMA was al-so significantly reduced.The algorithm can significantly improve the reconstruction performance and provide technical support for the re-search of brain image reconstruction.

Abstract:To address the N/2 Ghost artifacts caused by phase encoding errors in diffusion-weighted imaging(DWI)for low-field(<1 T)magnetic resonance imaging(MRI)systems,we proposed an approach integrating deep learning and optimized filtering to e-liminate Ghost artifacts and enhance image quality.Firstly,an AR-UNet model incorporating dense residual connection and attention gate mechanism was developed to achieve precise segmentation of craniocerebral anatomical structures through feature reuse and dynam-ic weight allocation.Then,an edge-constrained accelerated non-local means filtering(SCNLM)was used to improve the computation-al efficiency of the model.The results showed that the average Dice similarity coefficient,accuracy rate and specificity of the model reached 0.932 1,0.943 6 and 0.943 0,respectively.SCNLM could increase the computational efficiency of the traditional NLM algo-rithm by approximately 50%,while maintaining the peak signal-to-noise ratio of 29.50 dB and the structural similarity of 0.88.This research can effectively suppress Ghost artifacts in low-field MRI systems and significantly enhances image quality.

Abstract:In response to the problem of uncontrollable thickness of the photoacoustic conversion layer and complex preparation process of traditional photoacoustic transducers,we proposed a preparation method of a novel carbon nanotube polydimethylsiloxane(CNT-PDMS)photoacoustic transducers.By systematically studying the relationship between the thickness of the CNT-PDMS trans-ducer photoacoustic conversion layer and the sound pressure intensity,the structural parameters of the conversion layer were optimized.The test result of acoustic performance showed that the sound pressure intensity increased with the increase of the conversion layer thickness,and reached peak at a laser pulse repetition frequency of 4 Hz and a conversion layer thickness of 83.4 μm.The result of the ultrasound thrombolysis experiment showed that under the action of ultrasound generated by the photoacoustic transducer,the rate of blood clot dissolution increased initially and then decreased with time.Thrombolysis of 5 min could reduce the weight of blood clots by about 40%.This study provides a new technological approach for the application of photoacoustic transducers in the medical field.

Abstract:As a key component of traditional Chinese medicine,tuina has garnered increasing global attention due to its distinctive theoretical framework and notable clinical efficacy.With rapid advancements in robotics,artificial intelligence and biomechanics,tuina robots representing the integration of traditional medicine and modern technology are undergoing a critical transition from laboratory re-search to clinical application.This paper provides a systematic review of the developmental context of tuina robot technology,compre-hensively analyzes the current research status in areas such as mechanical structure design,intelligent control algorithms,clinical ap-plication outcomes and explores key technological breakthroughs and challenges,and looks forward to the future development,aiming to provide valuable references for researchers in related fields.

Abstract:To implement the"action for upgrading and innovating the localization of medical devices"specify proposed in the"Shandong Province Biomedical Industry Science and Technology Innovation Action Plan(2025-2027)".We conducted a comprehen-sive analysis of the current situation of the medical device industry in Shandong Province by visiting and investigating medical device enterprises and related organizations,and put forward relevant measures and suggestions for the high-quality development of the medi-cal device industry in Shandong Province to provide references for the upgrading of the medical device industry in Shandong Province.

Abstract:To meet the demand for mobile home pain relief treatment and achieve efficient utilization of medical resources.We integrated the Internet of Things and surface neuroelectric modulation technology to develop a portable Internet of Things surface neuroelectric modula-tion analgesia system.This system covered three levels:device,cloud platform and mobile terminal.After the device connected to the wireless access point(AP),it could synchronize the stimulus parameters to the cloud platform and mobile terminal.Meanwhile,the cloud platform and mobile terminal could issue commands to the device.The three parties worked together to achieve ultra-remote control and dynamic man-agement of the system.Experiments showed that this device could stably output the set electrical stimulation parameters,and its average error was all less than 4.1%.After electrical regulation analgesia,the mechanical tenderness threshold at the pain induction point of the subjects was significantly increased compared with that before analgesia.The portable Internet of Things surface neuroelectric modulation analgesia sys-tem can effectively enhance the effect of mobile home analgesia treatment and has high application value.

Abstract:To enhance the precise application of diffuse correlation spectroscopy(DCS)in medicine,we combined microscopic red blood cell motion imaging with DCS measurement to explore the connection between microscopic red blood cell motion parameters and macroscopic blood flow parameters.Firstly,the sheep blood samples were systematically analyzed through high-speed microscopic ima-ging and DCS.Then,the red blood cell movement parameters were obtained by continuous microscopic imaging and inter-frame corre-lation analysis of images,the mean square displacement(MSD)and diffusion coefficient of red blood cells were determined.Finally,the tissue blood flow index(BFI)was measured by DCS.The results showed that in the experiments of sheep blood samples at different dilution levels,with the increase of the dilution level,the diffusion coefficient of red blood cells in sheep blood increased.At the same time,the BFI measured by DCS showed an upward trend,indicating that the freedom of movement and diffusion coefficient of red blood cells have been significantly enhanced.Among them,at the moderate dilution level,the red blood cell diffusion coefficient was signifi-cantly positively correlated with the BFI measured by DCS(maximum r=0.63,P<0.001).This method can provide an experimental basis for understanding the mechanism of action between microcirculation disorders and ischemic stroke,cardiovascular diseases and the tumor microenvironment.

Abstract:To address the problem of blurred boundaries and low segmentation accuracy in pancreas segmentation,we proposed a pancreatic segmentation network MSW-HRNet.Firstly,by integrating depthwise separable convolution and spatial attention mecha-nisms,a multi-scale upsample block(MUB)was designed to restore the detailed information during the multi-scale upsampling process,enhance the segmentation ability for small-sized target regions.Then,the sliding-window attention block(Swin-Block)was fused to sense the global context information across scales,improve the discrimination ability of the model on lesion tissue and complex background,and enhance the performance of pancreatic boundary segmentation under complex structures.Experimental results demon-strated that the Dice coefficient of this method reached 82.11%on the NIH public dataset and 86.93%on the private pancreatitis data-set,outperforming mainstream segmentation models,confirming its superiority and practicality in handling complex pathological mor-phologies.

Abstract:Optical coherence tomography(OCT)is a non-invasive high-resolution imaging technique and serves as the gold standard for diagnosis of retinal diseases.With the advancement of technology,OCT has achieved rapid development in imaging speed,resolution,and other aspects.It enables the detection of indicators such as retinal angiography,blood oxygen saturation,and blood flow velocity,and when combined with deep learning technology,provides important reference value for the diagnosis and treatment of oph-thalmic diseases.This article introduces the research progress of OCT in the field of ophthalmology from the perspectives of OCT struc-tural imaging and functional imaging,as well as the OCT image processing methods based on deep learning and looks forward to the ap-plication prospects of OCT in ophthalmic clinical practice.