Alzheimer’s disease (AD), a common form of dementia, is a progressive neurodegenerative condition that necessitates early and accurate diagnosis to minimize the risk of misdiagnosis, treatment delays, or inappropriate interventions as it advances from the cognitively normal (CN) stage to mild cognitive impairment (MCI) and ultimately to AD. Interpreting diverse multimodal data for reliable diagnosis and progression monitoring typically depends on expert evaluation, which can be subjective. Moreover, existing automated methods often rely on rigid, discrete classifications that fail to account for uncertainty and transitional dynamics between stages. To address these limitations, the proposed method provides a comprehensive and structured input from multimodal data to generate a synthetic medical report as a supportive document with AI tool guidance in the absence of detailed analysis and handwritten reports. MRI and PET data are used to extract radiomic features and generate correlation maps at both whole-brain and sub-brain levels. This allows for the simultaneous assessment of structural atrophy and metabolic changes. Diagnostic classification into CN, MCI, and AD is performed using multiple machine learning models, with explicit identification of possible inter-stage and cross-modality conflicts. Additionally, principal component analysis (PCA) is applied to these correlation maps to detect longitudinal progression patterns, ranging from stable to aggressive trajectories, and to pinpoint the most affected brain regions. The resulting system not only enhances stage classification accuracy but also provides insight into disease evolution by leveraging the structural and functional relationships captured through radiomic correlations. These outputs, encompassing diagnostic insights, imaging-based findings, cognitive score trajectories, and conflict analysis, are structured into input data suitable for AI-guided synthetic ground-truth report generation due to the absence of handwritten data. The collection of the structured input and ground truth reports is used to fine-tune LLMs to obtain optimal and meaningful results. This approach enables the automatic production of detailed, personalized diagnostic reports. Ultimately, the integration of radiomic correlation mapping with PCA-driven progression analysis offers a robust, non-invasive, and automated decision-support tool for comprehensive AD evaluation and reporting.

Sensor-based sleep motion recognition (SMR) can effectively monitor human health and quality of life, but traditional methods have privacy and security risks. Therefore, this paper uses accelerometers and electrocardiograms (ECG) sensors to measure human sleep behavior data and proposes an SMR method based on the Fibonacci ladder search-multilayer extreme learning machine (FLS-MLELM). Firstly, an improved quantum genetic algorithm and a multiple bidirectional long short-term memory network are utilized to extract ECG-heart rate variability signals and motion acceleration signals from human sleep, respectively. Secondly, the FLS algorithm is employed to optimize the hyperparameters of MLELM, thereby improving the recognition accuracy of the model. Finally, in these two scenarios, the recognition accuracy of this method reached 95.54% (Chinese PLA General Hospital-SMR dataset) and 95.07% (self-collected SMR datasets), improving by 1.59%-25.72% and 2.31%-24.95%, respectively, compared to traditional algorithms. This proves that the proposed recognition method has potential application prospects in smart sleep-monitoring systems.

The objective is to address the issues of data imbalance, overfitting, and inadequate generalization ability in skin disease datasets and recognition models. The proposed model for the classification of skin diseases is based on the fusion of features and the utilization of transfer learning. The model's architecture is predicated on a dense connection network that serves as its fundamental framework, with LBP and HOG features incorporated as supplementary inputs. Subsequently, a feature fusion module integrated with an attention mechanism is employed to extract and combine features. Finally, the Softmax-loss function of category equilibrium and the domain adaptive strategy based on the maximum mean difference are established. The integration of prior knowledge into the deep network is a critical step in addressing the challenges of overfitting and data imbalance in skin disease classification. The FFTL-Net achieved AUC value of 98.16% on the International skin imaging collaboration (ISIC) 2018 dataset and 98.31% on the ISIC 2019 dataset. This represents an improvement of 1.25% and 0.33% compared to the second-ranked algorithm, respectively. The experimental results demonstrate the efficacy of the model in addressing the data imbalance issue in skin disease datasets, with prediction accuracies of at least 93% being achieved for BCC and other rare samples. The model demonstrates superior recognition accuracy, augmented generalization capability, and an absence of indications of overfitting.

The precise subtyping of lung cancer remains a significant and challenging task in clinical practice, and existing computer-aided diagnostic systems often overlook complex and specialized medical knowledge. In response to these challenges, a Pathological Knowledge-inspired Multi-scale Transformer Network (PKMT-Net) was proposed for predicting lung cancer subtypes using histopathological images. PKMT-Net consists of three key modules: a multi-scale soft segmentation module, a cross-attention module, and a weighted multi-scale fusion module. Initially, the multi-scale soft segmentation module simulated the pathologist's reading of histopathological images at various scales, capturing both macroscopic and microscopic characteristics. This module implements a novel soft patch generation strategy to mitigate semantic information loss. Next, the cross-attention module, equipped with skip connections, emulated the pathologist's way of correlating macroscopic and microscopic tumor characteristics. Lastly, the weighted multi-scale fusion module modeled the pathologist's decision-making process by integrating macroscopic and microscopic characteristics. After iterative training, the PKMT-Net model delivered an outstanding performance, attaining Area Under the Curve (AUC) values of 0.9992 for the training set, 0.9959 for the validation set, and 0.9970 for an unseen test set. Compared to single-scale models, PKMTNet's AUC improved by at least 0.0210. The model's interpretability, clinical utility, as well as the outcomes of ablation studies were evaluated comprehensively. Furthermore, the PKMT-Net model's generalizability was demonstrated through additional datasets. These results underscore the feasibility and high performance of the PKMT-Net for the processing of histopathology images. The supporting codes of this work can be found at: https: //github.com/zzl2022/PKMT-Net.

The precise mutation prediction of the Epidermal Growth Factor Receptor (EGFR) holds paramount importance in clinical practice. Nevertheless, the persisting challenge lies in accurately conducting genomic profiling of lung cancer using a single biopsy sample, given the inherent tumor heterogeneity. To address this issue, an innovative approach using similarity-based multimodal data fuzzy fusion was presented to predict EGFR mutation. Initially, radiomics features were extracted from computerized tomography scans to quantitatively characterize tumors within the region of interest. Subsequently, three independent fundamental learners were trained based on preprocessed multimodal medical data. Once these fundamental learners generate membership degrees, fuzzy sets for EGFR genotyping were established. The Tanimoto coefficient was then employed to evaluate the similarity between the membership degrees of observed cases and ideal solutions. Ultimately, de-fuzzification through similarity ranking yielded a robust prediction for the EGFR mutation. The proposed multimodal medical data fuzzy fusion demonstrates promising predictive performance, achieving an area under curve value of 0.8878 in an independent test cohort. The proposed work has the potential to serve as a robust and intelligent decision-making system for clinicians.

To address the issues of redundant high-dimensional action features, few action behavior classifications, and weak generalization ability of recognition models in traditional human behavior recognition (HBR), this paper proposes an HBR method based on Nonlinear Shannon's Principal Component Analysis (NSPCA) and Multi-Strategy Improved Remora Optimization Algorithm (MSIROA)-Adaptive Bi-kernel Support Vector Machine (ABKSVM). First, the NSPCA is employed to extract features from multiple sources of information and address the issue of nonlinear features within the data. Then, the selected principal component fusion features are input into the MSIROA-ABKSVM model, to achieve recognition of human behaviors. By utilizing an improved SVM, the extracted features are recognized to enhance the ability to identify behavior accurately. The experimental results indicate that the cumulative variance contribution rate of the six principal components selected by the NSPCA method reaches 85 % to simplify the data structure. Using the analysis of performance indicators, the classification method achieved an accuracy of 99.3 % and 99.5 % on the self-collected HBR datasets and the Chinese PLA General Hospital (PLAGH)-HBR dataset, respectively, outperforming other state-of-the-art methods. The results show that the HBR model can identify 33 different human behaviors, providing a new method for improving the recognition rate and effectiveness of daily activity monitoring for the elderly.

To solve the issues of low estimation accuracy and limited applicability of the existing stride length estimation (SLE) methods in indoor personnel positioning, a novel mixed stride length estimation (MSLE) model is proposed, combining the adaptive Harris hawk optimization (AHHO)-backpropagation neural network (BPNN) and the inverted pendulum model. This model accurately estimates pedestrian stride-length by analyzing and extracting features from the accelerometer, gyroscope data, and surface electromyography (SEMG) signals. The collected sensor signals are preprocessed using the second-generation wavelet algorithm. The peak detection algorithm is employed for stride counting, and based on this, a SLE algorithm is proposed using an AHHO-BPNN model. Subsequently, a MSLE model is developed by fitting it with a three-dimensional linear inverted pendulum model (3D-LIPM). The resulting model is then tested for individual indoor SLE. The experimental results indicate that the MSLE model can accurately estimate indoor stride lengths under different walking speeds. Compared with traditional models, it has lower SLE errors, meeting the requirements of personal indoor positioning. Therefore, this model has great potential for applications in fields such as rehabilitation medicine and remote monitoring.

The data from the public dataset that support the findings of this study are openly available in [ISIC 2019] at https://challenge.isic-archive.com/data/. The remaining data that support the findings of this study are available from the corresponding author upon reasonable request.

Objective. Liver cancer is a major global health problem expected to increase by more than 55% by 2040. Accurate segmentation of liver tumors from computed tomography (CT) images is essential for diagnosis and treatment planning. However, this task is challenging due to the variations in liver size, the low contrast between tumor and normal tissue, and the noise in the images. Approach. In this study, we propose a novel method called location-related enhancement network (LRENet) which can enhance the contrast of liver lesions in CT images and facilitate their segmentation. LRENet consists of two steps: (1) locating the lesions and the surrounding tissues using a morphological approach and (2) enhancing the lesions and smoothing the other regions using a new loss function. Main results. We evaluated LRENet on two public datasets (LiTS and 3Dircadb01) and one dataset collected from a collaborative hospital (Liver cancer dateset), and compared it with state-of-the-art methods regarding several metrics. The results of the experiments showed that our proposed method outperformed the compared methods on three datasets in several metrics. We also trained the Swin-Transformer network on the enhanced datasets and showed that our method could improve the segmentation performance of both liver and lesions. Significance. Our method has potential applications in clinical diagnosis and treatment planning, as it can provide more reliable and informative CT images of liver tumors.

There are some related existed models of NLP such as Word Embedding, CRF, Deep Neural Networks et al. The simple and necessary introductions are listed as following.

This book explores the applications of cutting-edge technologies and presents several innovative methods in healthcare

In this chapter, a medical NER model, which is simply called as BERT-Attention-SCLSTM-CRF, is proposed by adding extended input units based on BiLSTM with attention distraction mechanism.

Finally, the innovative standards of smart medicine devices are presented, mainly including the domain information models of the personal health devices and their applications.

目的 探讨慢性阻塞性肺疾病(COPD)合并阿尔茨海默病(AD)患者住院期间发生肺部感染的危险因素及改良虚弱指数(mFI)对预后的评估价值.方法 回顾性选取解放军总医院第二医学中心神经内科COPD合并AD且发生了肺部感染的患者80例,设为感染组;选取COPD合并AD但未发生肺部感染的患者80例,设为非感染组.再根据患者在住院期间的存活或死亡情况,将160例患者分入死亡组(11例)和存活组(149例).收集研究所需要的各项信息,包括N末端B型脑钠肽前体(NT-proBNP)、超敏C-反应蛋白(hs-CRP)、降钙素原(PCT)水平等,计算mFI评分.采用Logistic回归分析探讨COPD合并AD患者肺部感染及预后的影响因素,绘制受试者工作特征曲线(ROC),计算ROC曲线下面积(AUC)、敏感度、特异度和最佳截断值.结果 感染组患者的吸烟构成比以及NT-proBNP、CysC、hs-CRP、PCT、mFI水平均高于非感染组患者(P＜0.05).死亡组患者NT-proBNP、CysC、hs-CRP、PCT、mFI水平均高于存活组患者(P＜0.0 5).Logistic回归分析结果显示,NT-proBNP、hs-CRP、mFI是COPD合并AD患者肺部感染、死亡的独立影响因素(P＜0.05).mFI评分评估COPD合并AD患者肺部感染与预后的灵敏度、特异度分别为80.06％、59.38％和85.65％、72.53％.结论 mFI是COPD合并AD患者出现肺部感染及预后情况的独立预测因素,具有较高的临床评估价值,可作为预测COPD合并AD患者住院期间出现肺部感染和预后情况的指标.

In recent years, the rapid development of computer technology has made great progress in the field of smart health-care. Using data mining technology can obtain useful information in medical big data, discover the correlations among diseases and achieve the prevention and control of diseases. However, the traditional mining algorithms can no longer satisfy the demands of medical big data, it is an important direction of future research to improve and optimize the algorithms to make them applicable to the medical field. Based on this, this paper improves Apriori algorithm through parallel processing, matrix compression, and the introduction of lifting rate and interest. The optimized algorithm is called CI-Apriori, and an association rule model of diabetes complications based on CI-Apriori is proposed. The experimental results show that CI-Apriori has a great improvement in time and space efficiency, and can mine the strong association rules among diabetes complications faster and more effectively, so as to find the medical laws for the prevention and treatment of diseases.

The second chapter introduces the status and prospects of blockchain combined with AI in the field of healthcare. The core technologies of blockchain and the blockchain-based AI system framework are respectively presented. Then, the main applications in the intelligent sharing of eletronic medical records and traceability of drugs are listed.

BACKGROUND:Growth differentiation factor 15 (GDF-15) has been explored as a potential biomarker for various inflammatory diseases and cardiovascular events. This study aimed to assess the predictive role of GDF-15 levels in cardiovascular events and all-cause mortality, considering traditional risk factors and other biomarkers.METHODS:A prospective study was conducted and 3699 patients with stable coronary artery disease (CAD) were enrolled into the research. Baseline GDF-15 levels were measured. Median follow-up was 3.1 years during the study. We analyzed clinical variables and several biomarkers. Multivariable Cox regression analysis was performed to evaluate prognostic performance of GDF-15 levels in predicting myocardial infarction (MI), heart failure, stroke, cardiovascular death, and non-cardiovascular death.RESULTS:Baseline GDF-15 levels for 3699 patients were grouped by quartile (≤ 1153, 1153-1888, 1888-3043, > 3043 ng/L). Higher GDF-15 levels were associated with older age, male gender, history of hypertension, and elevated levels of N-terminal pro B-type natriuretic peptide (NT-pro BNP), soluble suppression of tumorigenesis-2 (sST2), and creatine (each with P < 0.001). Adjusting for established risk factors and biomarkers in Cox proportional hazards models, a 1 standard deviation (SD) increase in GDF-15 was associated with elevated risk of clinical events [hazard ratio (HR) = 2.18, 95% confidence interval (CI): (1.52-3.11)], including: MI [HR = 2.83 95% CI: (1.03-7.74)], heart failure [HR = 2.71 95% CI: (1.18-6.23)], cardiovascular and non-cardiovascular death [HR = 2.48, 95% CI (1.49-4.11)] during the median follow up of 3.1 years.CONCLUSIONS:Higher levels of GDF-15 consistently provides prognostic information for cardiovascular events and all cause death, independent of clinical risk factors and other biomarkers. GDF-15 could be considered as a valuable addition to future risk prediction model in secondary prevention for predicting clinical events in patient with stable CAD.

Accurate segmentation of ground-glass opacity (GGO) is an important premise for doctors to judge COVID-19. Aiming at the problem of mis-segmentation for GGO segmentation methods, especially the problem of adhesive GGO connected with chest wall or blood vessel, this paper proposes an accurate segmentation of GGO based on fuzzy c-means (FCM) clustering and improved random walk algorithm. The innovation of this paper is to construct a Markov random field (MRF) with adaptive spatial information by using the spatial gravity Model and the spatial structural characteristics, which is introduced into the FCM model to automatically balance the insensitivity to noise and preserve the effectiveness of image edge details to improve the clustering accuracy of image. Then, the coordinate values of nodes and seed points in the image are combined with the spatial distance, and the geodesic distance is added to redefine the weight. According to the edge density of the image, the weight of the grayscale and the spatial feature in the weight function is adaptively calculated. In order to reduce the influence of edge noise on GGO segmentation, an adaptive snowfall model is proposed to preprocess the image, which can suppress the noise without losing the edge information. In this paper, CT images of different types of COVID-19 are selected for segmentation experiments, and the experimental results are compared with the traditional segmentation methods and several SOTA methods. The results suggest that the paper method can be used for the auxiliary diagnosis of COVID-19, so as to improve the work efficiency of doctors.

Aiming at the problem that the single CT image signal feature recognition method in the self-diagnosis of diseases cannot accurately and reliably classify COVID-19, and it is easily confused with suspected cases. The collected CT signals and experimental indexes are extracted to construct different feature vectors. The support vector machine is optimized by the improved whale algorithm for the preliminary diagnosis of COVID-19, and the basic probability distribution function of each evidence is calculated by the posterior probability modeling method. Then the similarity measure is introduced to optimize the basic probability distribution function. Finally, the multi-domain feature fusion prediction model is established by using the weighted D-S evidence theory. The experimental results show that the fusion of multi-domain feature information by whale optimized support vector machine and improved D-S evidence theory can effectively improve the accuracy and the precision of COVID-19 autonomous diagnosis. The method of replacing a single feature parameter with multi-modal indicators (CT, routine laboratory indexes, serum cytokines and chemokines) provides a more reliable signal source for the diagnosis model, which can effectively distinguish COVID-19 from the suspected cases.

The quality of asymptomatic corona virus disease 2019 (COVID-19) computed tomography (CT) image is reduced due to interference from Gaussian noise, which affects the subsequent image processing. Aiming at the problem that asymptomatic COVID-19 CT image often have small flake ground-glass shadow in the early lesions, and the density is low, which is easily confused with noise. A denoising method of wavelet transform with shrinkage factor is proposed. The threshold decreases with the increase of decomposition scale, and it reduces the misjudgment of signal points. In the advanced stage, the range of lesions increases, with consolidation and fibrosis in different sizes, which have similar gray value to the CT images of suspected cases. Aiming at the problems of low contrast and fuzzy boundary in the traditional wavelet transform, the threshold function based on the optimization of parameters combined with the improved particle swam optimization (PSO) is proposed, so that the parameters of wavelet threshold function can change adaptively according to the lung lobe and ground-glass lesions with fewer iterations. The simulation results show that the paper method is significantly better than other algorithms in peak signal-to-noise ratio (PSNR), signal-to-noise ratio (SNR) and mean absolute error (MSE). For example, aiming at the early asymptomatic COVID-19, compared with the comparison methods, the PSNR under the proposed method has increased by about 5 dB, the MSE has been greatly reduced, and the SNR has increased by about 6.1 dB. It can be seen that the denoising effect under the proposed method is the best.