Performance Of Model Research Articles

Machine Learning Prediction of Prediabetes in a Young Male Chinese Cohort with 5.8-Year Follow-Up

The identification of risk factors for future prediabetes in young men remains largely unexamined. This study enrolled 6247 young ethnic Chinese men with normal fasting plasma glucose at the baseline (FPGbase), and used machine learning (Mach-L) methods to predict prediabetes after 5.8 years. The study seeks to achieve the following: 1. Evaluate whether Mach-L outperformed traditional multiple linear regression (MLR). 2. Identify the most important risk factors. The baseline data included demographic, biochemistry, and lifestyle information. Two models were built, where Model 1 included all variables and Model 2 excluded FPGbase, since it had the most profound effect on prediction. Random forest, stochastic gradient boosting, eXtreme gradient boosting, and elastic net were used, and the model performance was compared using different error metrics. All the Mach-L errors were smaller than those for MLR, thus Mach-L provided the most accurate results. In descending order of importance, the key factors for Model 1 were FPGbase, body fat (BF), creatinine (Cr), thyroid stimulating hormone (TSH), WBC, and age, while those for Model 2 were BF, white blood cell, age, TSH, TG, and LDL-C. We concluded that FPGbase was the most important factor to predict future prediabetes. However, after removing FPGbase, WBC, TSH, BF, HDL-C, and age were the key factors after 5.8 years.

Prediction of Air Quality Data with Feature Selection and Machine Learning Algorithms

Predicting air quality is essential for managing public health and the environment. This study looks into how well machine learning (ML) algorithms forecast air quality data and incorporates feature selection strategies for better model performance. Air pollution is one of the biggest issues facing humanity today. Every year, air pollution causes millions of deaths due to direct or indirect effects. These days, there is a critical need for effective measures to mitigate the negative impacts of air pollution. Usually, the answers to the issues with air pollution are hasty decisions that don't end up helping. It is vital to concentrate efforts on the pollutants that cause the majority of air pollution in order to create an efficient counter-strategy. In many industrial and urban regions today, monitoring and safeguarding air quality has emerged as one of the government's top priorities. The combustion of fossil fuels, traffic and weather conditions, and industrial parameters

Bayesian network analysis of factors influencing type 2 diabetes, coronary heart disease, and their comorbidities

ObjectiveBayesian network (BN) models were developed to explore the specific relationships between influencing factors and type 2 diabetes mellitus (T2DM), coronary heart disease (CAD), and their comorbidities. The aim was to predict disease occurrence and diagnose etiology using these models, thereby informing the development of effective prevention and control strategies for T2DM, CAD, and their comorbidities.MethodEmploying a case-control design, the study compared individuals with T2DM, CAD, and their comorbidities (case group) with healthy counterparts (control group). Univariate and multivariate Logistic regression analyses were conducted to identify disease-influencing factors. The BN structure was learned using the Tabu search algorithm, with parameter estimation achieved through maximum likelihood estimation. The predictive performance of the BN model was assessed using the confusion matrix, and Netica software was utilized for visual prediction and diagnosis.ResultThe study involved 3,824 participants, including 1,175 controls, 1,163 T2DM cases, 982 CAD cases, and 504 comorbidity cases. The BN model unveiled factors directly and indirectly impacting T2DM, such as age, region, education level, and family history (FH). Variables like exercise, LDL-C, TC, fruit, and sweet food intake exhibited direct effects, while smoking, alcohol consumption, occupation, heart rate, HDL-C, meat, and staple food intake had indirect effects. Similarly, for CAD, factors with direct and indirect effects included age, smoking, SBP, exercise, meat, and fruit intake, while sleeping time and heart rate showed direct effects. Regarding T2DM and CAD comorbidities, age, FBG, SBP, fruit, and sweet intake demonstrated both direct and indirect effects, whereas exercise and HDL-C exhibited direct effects, and region, education level, DBP, and TC showed indirect effects.ConclusionThe BN model constructed using the Tabu search algorithm showcased robust predictive performance, reliability, and applicability in forecasting disease probabilities for T2DM, CAD, and their comorbidities. These findings offer valuable insights for enhancing prevention and control strategies and exploring the application of BN in predicting and diagnosing chronic diseases.

Evaluation of influence of wall roughness on wall turbulence using LES simulation with MSGD model for liquid Li flow inside two-staged contraction nozzle

ABSTRACT Liquid lithium (Li) jet is planned as a beam target in fusion neutron sources (FNSs), such as IFMIF, A-FNS and IFMIF-DONES. For the safety and the efficiency of such FNSs, the high-speed Li jet need to be kept stable. For investigation of the detailed flow pattern inside the Li jet, numerical approaches using CFD simulation have been required to evaluate it because of its opacity. As one of issues about the inner flow structure, the influence of wall roughness on the vortex structure and surface variation is concerned in long-term operation of actual FNSs. In our previous simulation, in order to consider wall roughness in LES simulation, MSGD model was adopted to ANSYS Fluent using User-Defined Function, and CFD simulation in a simple parallel plate flow was conducted to confirm the performance of MSGD model. In this study, firstly, MSGD model was verified by LES simulation of Li flowing on smooth and rough plates. After that, LES simulation with MSGD model for two-staged contraction nozzle, installed into Li loop of Osaka University, was conducted. It was found that MSGD model was valid and spectrum of velocity in roughness wall cases clearly had different behavior from that in smooth wall case.

Assessment of cerebral venous sinus: Anatomical and functional diagnostic performance of three-dimensional reconstruction models based on venous sinus MRI and CT images

Venous sinus stenosis is commonly observed in patients presenting with pulsatile tinnitus (PT). While magnetic resonance imaging (MRI) and computed tomography (CT) are commonly used for assessing venous sinus geometries, the preferred modality remains unclear. In this study, we reconstructed the three-dimensional (3D) geometries of the venous sinus using MRI and CT imaging data from 20 PT patients. We conducted comparisons of the anatomical features of the venous sinus through case-wise analysis and anatomic geometrical parameter-wise analysis. Our findings indicate that by taking the geometries from CT as a reference, MRI could provide a better illustration of venous structure, primarily due to a stronger flow signal concentrated in the vascular tree. We observed high agreements in anatomic parameters measured from 3D geometries reconstructed based on CT and MRI in 19 out of 20 cases. Notably, the cross-sectional area of the sinus and segment length displayed the highest consistency, with a mean difference of -5.01% and 6.5% between modalities, respectively. In addition, we noticed that 55% of cases exhibited consistency in analyzing the confluence of the sinus, while variants of connectivity and collateral branching were observed between CT and MRI. Importantly, CT-based geometric reconstruction provided better detail of inflow side branches in the straight sinus, whereas MRI preserved more side branches of outflow in the downstream sinus. It is important to note that CT-based evaluation may be affected by the bone structures surrounding the venous sinus, whereas MRI-based evaluation focuses on blood flow to the segments, potentially indicating both anatomical and functional abnormalities.

No-Reference Hyperspectral Image Quality Assessment via Ranking Feature Learning

In hyperspectral image (HSI) reconstruction tasks, due to the lack of ground truth in real imaging processes, models are usually trained and validated on simulation datasets and then tested on real measurements captured by real HSI imaging systems. However, due to the gap between the simulation imaging process and the real imaging process, the best model validated on the simulation dataset may fail on real measurements. To obtain the best model for the real-world task, it is crucial to design a suitable no-reference HSI quality assessment metric to reflect the reconstruction performance of different models. In this paper, we propose a novel no-reference HSI quality assessment metric via ranking feature learning (R-NHSIQA), which calculates the Wasserstein distance between the distribution of the deep features of the reconstructed HSIs and the benchmark distribution. Additionally, by introducing the spectral self-attention mechanism, we propose a Spectral Transformer (S-Transformer) to extract the spatial-spectral representative deep features of HSIs. Furthermore, to extract quality-sensitive deep features, we use quality ranking as a pre-training task to enhance the representation capability of the S-Transformer. Finally, we introduce the Wasserstein distance to measure the distance between the distribution of the deep features and the benchmark distribution, improving the assessment capacity of our method, even with non-overlapping distributions. The experimental results demonstrate that the proposed metric yields consistent results with multiple full-reference image quality assessment (FR-IQA) metrics, validating the idea that the proposed metric can serve as a substitute for FR-IQA metrics in real-world tasks.

Multi-strategy enhanced snake optimizer for quantitative structure-activity relationship modeling

Quantitative structure-activity relationship modeling techniques are widely employed in diverse areas of chemistry, including environmental risk assessment of compounds and virtual screening of drugs. The modeling process entails optimizing the hyper parameters of the model and selecting appropriate molecular descriptors. It is a typical optimization problem with high-dimensional, nonlinear, continuous and discrete variables coexisting. To address this challenge, this paper proposes an enhanced snake optimizer with multiple strategies. First, new food emergence and temperature change mechanisms are introduced to balance the exploration and exploitation. Second, a death and regeneration mechanism is employed to enhance the algorithm's ability to escape from local optima. Third, new formulas for food searching, fighting, mating and eating are designed to simplify the complexity of the algorithm and improve its search efficiency and accuracy. Fourth, the performance of the proposed algorithm is evaluated using the CEC2017 benchmark functions, eight engineering cases with continuous variables, and six feature selection cases with discrete variables. Experimental results demonstrate that the algorithm significantly outperforms comparison algorithms. Finally, the algorithm was applied to quantitative structure-activity relationship modeling for environmental toxicity prediction. Compared with other algorithms, the algorithm can better configure hyper-parameters and select fewer and more effective features for environmental toxicity prediction model. It will be an effective tool to improve the performance of quantitative structure-activity relationship models.

Predictive value of the domain specific PLA2R antibodies for clinical remission in patients with primary membranous nephropathy: A retrospective study.

M-type phospholipase A2 receptor (PLA2R) is a major auto-antigen of primary membranous nephropathy(PMN). Anti-PLA2R antibody levels are closely associated with disease severity and therapeutic effectiveness. Analysis of PLA2R antigen epitope reactivity may have a greater predictive value for remission compared with total PLA2R-antibody level. This study aims to elucidate the relationship between domain-specific antibody levels and clinical outcomes of PMN. This retrospective analysis included 87 patients with PLA2R-associated PMN. Among them, 40 and 47 were treated with rituximab (RTX) and cyclophosphamide (CTX) regimen, respectively. The quantitative detection of -immunoglobulin G (IgG)/-IgG4 targeting PLA2R and its epitope levels in the serum of patients with PMN were obtained through time-resolved fluorescence immunoassays and served as biomarkers in evaluating the treatment effectiveness. A predictive PMN remission possibility nomogram was developed using multivariate logistic regression analysis. Discrimination in the prediction model was assessed using the area under the receiver operating characteristic curve (AUC-ROC).Bootstrap ROC was used to evaluate the performance of the prediction model. After a 6-month treatment period, the remission rates of proteinuria, including complete remission and partial remission in the RTX and CTX groups, were 70% and 70.21% (P = 0.983), respectively. However, there was a significant difference in immunological remission in the PLA2R-IgG4 between the RTX and CTX groups (21.43% vs. 61.90%, P = 0.019). Furthermore, we found differences in PLA2R-CysR-IgG4(P = 0.030), PLA2R-CTLD1-IgG4(P = 0.005), PLA2R-CTLD678-IgG4(P = 0.003), and epitope spreading (P = 0.023) between responders and non-responders in the CTX group. Multivariate logistic analysis showed that higher levels of urinary protein (odds ratio [OR], 0.49; 95% confidence interval [CI], 0.26-0.95; P = 0.035) and higher levels of PLA2R-CTLD1-IgG4 (OR, 0.79; 95%CI,0.62-0.99; P = 0.041) were independent risk factors for early remission. A multivariate model for estimating the possibility of early remission in patients with PMN is presented as a nomogram. The AUC-ROC of our model was 0.721 (95%CI, 0.601-0.840), in consistency with the results obtained with internal validation, for which the AUC-ROC was 0.711 (95%CI, 0.587-0.824), thus, demonstrating robustness. Cyclophosphamide can induce immunological remission earlier than rituximab at the span of 6 months. The PLA2R-CTLD1-IgG4 has a better predict value than total PLA2R-IgG for remission of proteinuria at the 6th month.

Association between wrist-worn free-living accelerometry and hand grip strength in middle-aged and older adults

IntroductionWrist-worn activity monitors have seen widespread adoption in recent times, particularly in young and sport-oriented cohorts, while their usage among older adults has remained relatively low. The main limitations are in regards to the lack of medical insights that current mainstream activity trackers can provide to older subjects. One of the most important research areas under investigation currently is the possibility of extrapolating clinical information from these wearable devices.MethodsThe research question of this study is understanding whether accelerometry data collected for 7-days in free-living environments using a consumer-based wristband device, in conjunction with data-driven machine learning algorithms, is able to predict hand grip strength and possible conditions categorized by hand grip strength in a general population consisting of middle-aged and older adults.ResultsThe results of the regression analysis reveal that the performance of the developed models is notably superior to a simple mean-predicting dummy regressor. While the improvement in absolute terms may appear modest, the mean absolute error (6.32 kg for males and 4.53 kg for females) falls within the range considered sufficiently accurate for grip strength estimation. The classification models, instead, excel in categorizing individuals as frail/pre-frail, or healthy, depending on the T-score levels applied for frailty/pre-frailty definition. While cut-off values for frailty vary, the results suggest that the models can moderately detect characteristics associated with frailty (AUC-ROC: 0.70 for males, and 0.76 for females) and viably detect characteristics associated with frailty/pre-frailty (AUC-ROC: 0.86 for males, and 0.87 for females).Conclusions The results of this study can enable the adoption of wearable devices as an efficient tool for clinical assessment in older adults with multimorbidities, improving and advancing integrated care, diagnosis and early screening of a number of widespread diseases.

A Systematic Review on Deep Learning Model in Computer-aided Diagnosis for Anterior Cruciate Ligament Injury.

In developing Computer-Aided Diagnosis (CAD), a Convolutional Neural Network (CNN) has been commonly used as a Deep Learning (DL) model. Although it is still early, DL has excellent potential in implementing computers in medical diagnosis. This study reviews the use of DL for Anterior Cruciate Ligament (ACL) tear diagnosis. A comprehensive search was performed in PubMed, Embase, and Web of Science databases from 2018 to 2024. The included study criteria used MRI images to evaluate ACL tears, and the diagnosis of ACL tears was performed using the DL model. We summarized the paper by reporting their model accuracy, model comparison with arthroscopy, and explainable. AI implementation in tabular format; we conclude that many medical professionals believe that arthroscopic diagnosis is the most reliable method for diagnosing ACL tears. However, due to its intrusive treatment, CAD is projected to be able to produce similar outcomes from MRI scan results. To gain the trust of physicians and meet the demand for reliable knee injury detection systems, an algorithm for CAD should also meet several criteria, such as being transparent, interpretable, explainable, and easy to use. Therefore, future works should consider creating an Explainable DL model for ACL tear diagnosis. It is also essential to evaluate the performance of this Explainable DL model compared to the gold standard of arthroscopy diagnosis. There are issues regarding the need for Explainable DL in CAD to increase confidence in its result while also highlighting the importance of the involvement of medical practitioners in system design. There is no funding for this work.

TM-Score predicts immunotherapy efficacy and improves the performance of the machine learning prognostic model in gastric cancer

Immunotherapy is becoming increasingly important, but the overall response rate is relatively low in the treatment of gastric cancer (GC). The application of tumor mutational burden (TMB) in predicting immunotherapy efficacy in GC patients is limited and controversial, emphasizing the importance of optimizing TMB-based patient selection. By combining TMB and major histocompatibility complex (MHC) related hub genes, we established a novel TM-Score. This score showed superior performance for immunotherapeutic selection (AUC = 0.808) compared to TMB, MSI status, and EBV status. Additionally, it predicted the prognosis of GC patients. Subsequently, a machine learning model adjusted by the TM-Score further improved the accuracy of survival prediction (AUC > 0.8). Meanwhile, we found that GC patients with low TM-Score had a higher mutation frequency, higher expression of HLA genes and immune checkpoint genes, and higher infiltration of CD8+ T cells, CD4+ helper T cells, and M1 macrophages. This suggests that TM-Score is significantly associated with tumor immunogenicity and tumor immune environment. Notably, based on the RNA-seq and scRNA-seq, it was found that AKAP5, a key component gene of TM-Score, is involved in anti-tumor immunity by promoting the infiltration of CD4+ T cells, NK cells, and myeloid cells. Additionally, siAKAP5 significantly reduced MHC-II mRNA expression in the GC cell line. In addition, our immunohistochemistry assays confirmed a positive correlation between AKAP5 and human leukocyte antigen (HLA) expression. Furthermore, AKAP5 levels were higher in patients with longer survival and those who responded to immunotherapy in GC, indicating its potential value in predicting prognosis and immunotherapy outcomes. In conclusion, TM-Score, as an optimization of TMB, is a more precise biomarker for predicting the immunotherapy efficacy of the GC population. Additionally, AKAP5 shows promise as a therapeutic target for GC.

Impact of Transfer Learning Using Local Data on Performance of a Deep Learning Model for Screening Mammography.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To investigate the issues of generalizability and replication of deep learning (DL) models by assessing performance of a screening mammography DL system developed at New York University (NYU) on a local Australian dataset. Materials and Methods In this retrospective study, all individuals with biopsy and surgical pathology-proven lesions and age-matched controls were identified from a South Australian public mammography screening program (January 2010 to December 2016). The primary outcome was DL system performance, measured with the area under the receiver operating characteristic curve (AUC), in classifying invasive breast cancer or ductal carcinoma in situ (n = 425) from no malignancy (n = 490) or benign lesions (n = 44) in age-matched controls. The NYU system, including models without (NYU1) and with (NYU2) heatmaps, was tested in its original form, after training from scratch (without transfer learning; TL), after retraining with TL. Results The local test set comprised 959 individuals (mean age, 62.5 years [SD, 8.5]; all female). The original AUCs for the NYU1 and NYU2 models were 0.83 (95%CI = 0.82-0.84) and 0.89 (95%CI = 0.88-0.89), respectively. When applied in their original form to the local test set, the AUCs were 0.76 (95%CI = 0.73-0.79) and 0.84 (95%CI = 0.82-0.87), respectively. After local training without TL, the AUCs were 0.66 (95%CI = 0.62-0.69) and 0.86 (95%CI = 0.84-0.88). After retraining with TL, the AUCs were 0.82 (95%CI = 0.80-0.85) and 0.86 (95%CI = 0.84-0.88). Conclusion A deep learning system developed using a U.S. dataset showed reduced performance when applied 'out of the box' to an Australian dataset. Local retraining with transfer learning using available model weights improved model performance. ©RSNA, 2024.

Comparative performance of intensive care mortality prediction models based on manually curated versus automatically extracted electronic health record data

IntroductionBenchmarking intensive care units for audit and feedback is frequently based on comparing actual mortality versus predicted mortality. Traditionally, mortality prediction models rely on a limited number of input variables and significant manual data entry and curation. Using automatically extracted electronic health record data may be a promising alternative. However, adequate data on comparative performance between these approaches is currently lacking. MethodsThe AmsterdamUMCdb intensive care database was used to construct a baseline APACHE IV in-hospital mortality model based on data typically available through manual data curation. Subsequently, new in-hospital mortality models were systematically developed and evaluated. New models differed with respect to the extent of automatic variable extraction, classification method, recalibration usage and the size of collection window. ResultsA total of 13 models were developed based on data from 5,077 admissions divided into a train (80%) and test (20%) cohort. Adding variables or extending collection windows only marginally improved discrimination and calibration. An XGBoost model using only automatically extracted variables, and therefore no acute or chronic diagnoses, was the best performing automated model with an AUC of 0.89 and a Brier score of 0.10. DiscussionPerformance of intensive care mortality prediction models based on manually curated versus automatically extracted electronic health record data is similar. Importantly, our results suggest that variables typically requiring manual curation, such as diagnosis at admission and comorbidities, may not be necessary for accurate mortality prediction. These proof-of-concept results require replication using multi-centre data.

Evaluating Polarizable Biomembrane Simulations against Experiments.

Owing to the increase of available computational capabilities and the potential for providing a more accurate description, polarizable molecular dynamics force fields are gaining popularity in modeling biomolecular systems. It is, however, crucial to evaluate how much precision is truly gained with increasing cost and complexity of the simulation. Here, we leverage the NMRlipids open collaboration and Databank to assess the performance of available polarizable lipid models─the CHARMM-Drude and the AMOEBA-based parameters─against high-fidelity experimental data and compare them to the top-performing nonpolarizable models. While some improvement in the description of ion binding to membranes is observed in the most recent CHARMM-Drude parameters, and the conformational dynamics of AMOEBA-based parameters are excellent, the best nonpolarizable models tend to outperform their polarizable counterparts for each property we explored. The identified shortcomings range from inaccuracies in describing the conformational space of lipids to excessively slow conformational dynamics. Our results provide valuable insights for the further refinement of polarizable lipid force fields and for selecting the best simulation parameters for specific applications.

Interpretable Landslide Susceptibility Evaluation Based on Model Optimization

Machine learning (ML) is increasingly utilized in Landslide Susceptibility Mapping (LSM), though challenges remain in interpreting the predictions of ML models. To reveal the response relationship between landslide susceptibility and evaluation factors, an interpretability model was constructed to analyze how the results of the ML model are realized. This study focuses on Zhenba County in Shaanxi Province, China, employing both Random Forest (RF) and Support Vector Machine (SVM) to develop LSM models optimized through Random Search (RS). To enhance interpretability, the study incorporates techniques such as Partial Dependence Plot (PDP), Local Interpretable Model-Agnostic Explanations (LIMEs), and Shapley Additive Explanations (SHAP). The RS-optimized RF model demonstrated superior performance, achieving an Area Under the Curve (AUC) of 0.965. The interpretability model identified the NDVI and distance from road as important factors influencing landslides occurrence. NDVI plays a positive role in the occurrence of landslides in this region, and the landslide-prone areas are within 500 m from the road. These analyses indicate the importance of improved hyperparameter selection in enhancing model accuracy and performance. The interpretability model provides valuable insights into LSM, facilitating a deeper understanding of landslide formation mechanisms and guiding the formulation of effective prevention and control strategies.

Suppressing label noise in medical image classification using mixup attention and self-supervised learning

Deep neural networks (DNNs) have been widely applied in medical image classification and achieve remarkable classification performance. These achievements heavily depend on large-scale accurately annotated training data. However, label noise is inevitably introduced in the medical image annotation, as the labeling process heavily relies on the expertise and experience of annotators. Meanwhile, DNNs suffer from overfitting noisy labels, degrading the performance of models. Therefore, in this work, we innovatively devise a noise-robust training approach to mitigate the adverse effects of noisy labels in medical image classification. Specifically, we incorporate contrastive learning and intra-group mixup attention strategies into vanilla supervised learning. The contrastive learning for feature extractor helps to enhance visual representation of DNNs. The intra-group mixup attention module constructs groups and assigns self-attention weights for group-wise samples, and subsequently interpolates massive noisy-suppressed samples through weighted mixup operation. We conduct comparative experiments on both synthetic and real-world noisy medical datasets under various noise levels. Rigorous experiments validate that our noise-robust method with contrastive learning and mixup attention can effectively handle with label noise, and is superior to state-of-the-art methods. An ablation study also shows that both components contribute to boost model performance. The proposed method demonstrates its capability of curb label noise and has certain potential toward real-world clinic applications.

A constrained maximum likelihood approach to developing well-calibrated models for predicting binary outcomes.

The added value of candidate predictors for risk modeling is routinely evaluated by comparing the performance of models with or without including candidate predictors. Such comparison is most meaningful when the estimated risk by the two models are both unbiased in the target population. Very often data for candidate predictors are sourced from nonrepresentative convenience samples. Updating the base model using the study data without acknowledging the discrepancy between the underlying distribution of the study data and that in the target population can lead to biased risk estimates and therefore an unfair evaluation of candidate predictors. To address this issue assuming access to a well-calibrated base model, we propose a semiparametric method for model fitting that enforces good calibration. The central idea is to calibrate the fitted model against the base model by enforcing suitable constraints in maximizing the likelihood function. This approach enables unbiased assessment of model improvement offered by candidate predictors without requiring a representative sample from the target population, thus overcoming a significant practical challenge. We study theoretical properties for model parameter estimates, and demonstrate improvement in model calibration via extensive simulation studies. Finally, we apply the proposed method to data extracted from Penn Medicine Biobank to inform the added value of breast density for breast cancer risk assessment in the Caucasian woman population.

SkinSwinViT: A Lightweight Transformer-Based Method for Multiclass Skin Lesion Classification with Enhanced Generalization Capabilities

In recent decades, skin cancer has emerged as a significant global health concern, demanding timely detection and effective therapeutic interventions. Automated image classification via computational algorithms holds substantial promise in significantly improving the efficacy of clinical diagnoses. This study is committed to mitigating the challenge of diagnostic accuracy in the classification of multiclass skin lesions. This endeavor is inherently formidable owing to the resemblances among various lesions and the constraints associated with extracting precise global and local image features within diverse dimensional spaces using conventional convolutional neural network methodologies. Consequently, this study introduces the SkinSwinViT methodology for skin lesion classification, a pioneering model grounded in the Swin Transformer framework featuring a global attention mechanism. Leveraging the inherent cross-window attention mechanism within the Swin Transformer architecture, the model adeptly captures local features and interdependencies within skin lesion images while additionally incorporating a global self-attention mechanism to discern overarching features and contextual information effectively. The evaluation of the model’s performance involved the ISIC2018 challenge dataset. Furthermore, data augmentation techniques augmented training dataset size and enhanced model performance. Experimental results highlight the superiority of the SkinSwinViT method, achieving notable metrics of accuracy, recall, precision, specificity, and F1 score at 97.88%, 97.55%, 97.83%, 99.36%, and 97.79%, respectively.

Fall-risk assessment of aged workers using wearable inertial measurement units based on machine learning

Falls are the major cause of fatal and nonfatal incidents among aged workers. Estimating individual fall risk is crucial for occupational protection and public health. This study aimed to develop a fall-risk assessment method based on Machine Learning (ML) and Inertial Measurement Units (IMUs). A total of 28 aged workers (60 to 80 years old) participated in this study, recruited from community sanitation workers and janitors. They were categorized into high and low fall-risk groups based on functional gait assessment. Each participant performed 5 sets of motion experiments in the lab, covering 6 daily motions. We gathered their kinematic data with IMUs for training ML-based fall-risk assessment models. Five classic ML classifiers were optimized using grid search, and a Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) hybrid deep neural network was built to recognize walking motion from kinematic data for input into the fall-risk assessment model. The results demonstrated that ML classifiers trained with walking motion data exhibited superior performance. Cross-validation revealed that the optimized Random Forest classifier, with IMU on the right upper leg, reached 91.3 % accuracy. Additionally, the CNN-LSTM models were evaluated using F1-score, which can balance accuracy and coverage of model performance evaluation. The CNN-LSTM achieved a maximum F1-score of 95.18 %, proving their effectiveness in extracting walking motion data. These findings indicate that the developed IMU-based ML method exhibits excellent performance in assessing fall risk, potentially allowing aged workers to assess their fall risk before clinical diagnosis by embedding it in wearable IMUs.

One-Shot Federated Learning with Label Differential Privacy

Federated learning (FL) has emerged as an extremely effective strategy for dismantling data silos and has attracted significant interest from both industry and academia in recent years. However, existing iterative FL approaches often require a large number of communication rounds and struggle to perform well on unbalanced datasets. Furthermore, the increased complexity of networks makes the application of traditional differential privacy to protect client privacy expensive. In this context, the authors introduce FedGM: a method designed to reduce communication overhead and achieve outstanding results in non-IID scenarios. FedGM is capable of achieving considerable accuracy, even with a small privacy budget. Specifically, the authors devise a method to extract knowledge from each client’s data by creating a scaled-down but highly effective synthesized dataset that can perform similarly to the original data. Additionally, the authors propose an innovative approach to applying label differential privacy to protect the synthesized dataset. The authors demonstrate the superiority of the approach over traditional methods by requiring only one communication round and by testing using four classification datasets for evaluation. Furthermore, when comparing the model performance for clients using their method against traditional solutions, the authors find that the approach achieves significant accuracy and better privacy.