OBJECTIVES:This study investigates the feasibility of using photoplethysmography (PPG) signals for supporting mitral regurgitation (MR) assessment and monitoring using machine learning techniques. METHODS:MR occurs when blood flows backward from the left ventricle to the left atrium due to mitral valve insufficiency. Although echocardiography is the standard diagnostic method, it requires expert interpretation. Other techniques, such as magnetic resonance imaging, cardiac catheterization, and angiography, are costly and invasive. This study investigates the feasibility of using PPG signals for exploring MR-related signal patterns and supporting non-invasive monitoring via machine learning techniques. RESULTS:From 1,590 ten-second PPG recordings, 37 features were extracted. Following feature selection, 11 features were retained. Using 10-fold cross-validation, the proposed ensemble classifier achieved an accuracy of 91.07 %, sensitivity of 0.9462, specificity of 0.8765, AUC of 0.9113, and MCC of 0.8249 using 11 selected PPG features. The results indicate that PPG-derived features may capture signal characteristics associated with MR. CONCLUSIONS:These findings demonstrate the feasibility of PPG-based machine learning approaches for supporting accessible and non-invasive monitoring of MR-related cardiovascular patterns.
OBJECTIVES:Tibial plateau fractures (TPFs) are serious knee injuries frequently associated with high-impact sports such as Wushu. This study proposes a parallel analytical framework that integrates artificial intelligence (AI)-based radiographic diagnosis with computational biomechanics to investigate the risk factors and injury mechanisms of TPFs during the Wushu Tornado Kick (Xuanfengjiao). METHODS:An EfficientNet-B3 deep learning model was trained on MURA and Kaggle datasets for automated radiographic detection of TPFs, achieving high diagnostic accuracy. Concurrently, 3D pose estimation and spatial kinematics were utilized to extract biomechanical parameters, specifically Knee Flexion Angle (KFA) and vertical Ground Reaction Forces (GRF), from elite Wushu athletes. RESULTS:The kinematic data were mapped to predict structural stresses on the tibial plateau upon landing impact. Our dual-pathway approach demonstrates that increased vertical GRF combined with suboptimal KFA significantly elevates the risk of specific Schatzker-type fractures. The AI model successfully correlated these mechanically predicted injury patterns with radiographic evidence. CONCLUSIONS:This framework bridges the gap between predictive sports biomechanics and clinical radiographic diagnosis, offering a robust tool for injury prevention and targeted rehabilitation in Wushu athletes.
OBJECTIVES:This study aimed to develop and preliminarily evaluate a lightweight wearable hand exoskeleton integrated with virtual reality (VR) feedback for post-stroke hand rehabilitation. METHODS:A compact exoskeleton with multi-link finger assistance, flex sensors, surface electromyography (EMG), adaptive control, and Unity-based VR tasks was designed. Twelve post-stroke patients completed 30-min sessions, five days per week, for four weeks. Outcomes included MCP and PIP range of motion (ROM), object manipulation time, ARAT, FMA, ADL, safety, and satisfaction. RESULTS:MCP ROM improved by 33 ± 7° and PIP ROM by 29 ± 6°, while object manipulation time decreased by 28 ± 10 %. ARAT, FMA, and ADL improved by 11.2 ± 3.1, 8.7 ± 2.0, and 15.0 ± 10.4, respectively. No adverse events were reported, and satisfaction was high. CONCLUSIONS:The system showed feasibility, safety, usability, and preliminary functional benefits. Larger controlled studies with longer follow-up are needed to confirm clinical efficacy.
OBJECTIVES:Self-etching primers have simplified the process of direct bonding of dental resins and orthodontic brackets. The aim of this study was to evaluate the cytotoxic impact of the leached components to L-929 fibroblast cultures. METHODS:Transbond™ Plus self-etching primer (3M Unitek, 2724 South Peck Road, Monrovin CA 91016 USA) was applied to enamel pieces in three different conditions and the specimens were added to L-929 fibroblast cultures. The incubation time was 72 h and the cells were counted by flow cytometry. Statistics was carried out using descriptive and explorative data analyses. RESULTS:In direct contact assays, the self-etching primer without bracket exhibited mild to moderate cytotoxicity. The application of primer, composite and bracket showed no relevant decrease of cell viability. After seven days, the cytotoxic effects had largely disappeared. Overall, the combination of self-etching primer, composite, bracket exhibits no significant cytotoxicity during the observation period. CONCLUSIONS:When self-etching primers are used according to the recommended clinical protocols in orthodontics, the method is regarded as safe and the release of leachables seems to have no adverse clinical effect.
INTRODUCTION:Robotic technologies are increasingly used in dentistry to improve procedural precision and standardization. However, clinical implementation remains heterogeneous due to differences in autonomy, technological maturity, and available clinical evidence. This review evaluates current applications using a dual-perspective framework. CONTENT:A narrative review of the literature was conducted using PubMed/MEDLINE, Scopus, and Web of Science. English-language publications from 2004 onwards were considered and selected according to their relevance to robotic system design, clinical applications, autonomy levels, technical performance, and translational challenges in dentistry. SUMMARY:Robotic applications have expanded across surgical, therapeutic, educational, and preventive domains. Most systems operate at assistive or semi-autonomous levels (Level 1-2), with the strongest clinical evidence observed in implant dentistry, where improved placement accuracy has been reported. Clinical adoption remains limited by high costs, technical complexity, and insufficient long-term outcome data. OUTLOOK:Dental robotics has evolved into a clinically applicable component of digital dentistry, but routine clinical implementation remains heterogeneous across dental specialties. The proposed dual-perspective framework facilitates evaluation of robotic technologies and their clinical readiness, while further clinical validation is needed to support routine implementation.
OBJECTIVES:Autism spectrum disorder (ASD) involves subtle and heterogeneous brain-structural differences that are not usually visible on routine structural magnetic resonance imaging (sMRI). This study investigated a segmentation-guided sMRI feature-classification framework for childhood ASD analysis. METHODS:Pediatric T1-weighted sMRI scans were obtained from the publicly available Autism Brain Imaging Data Exchange II (ABIDE-II) dataset. Preprocessing used the FMRIB Software Library (FSL) Brain Extraction Tool (BET), N4 bias-field correction, intensity normalization, and Advanced Normalization Tools (ANTs) registration. Gray matter, white matter, and cerebrospinal fluid (CSF) were segmented using FSL FAST, and cortical reconstruction/parcellation was performed using FreeSurfer recon-all. Extracted features included global structural summaries, regional cortical gray-matter volume, cortical surface area, cortical thickness, cortical-thickness variability, mean curvature, and folding index. Each participant was represented by a 414-dimensional numerical feature vector. Features were normalized, filtered, selected using minimum-redundancy maximum-relevance (mRMR) analysis, and classified using support vector machine, random forest, k-nearest neighbors, and multilayer perceptron models. RESULTS:The multilayer perceptron achieved 69.4 % accuracy, 67.9 % precision, 70.0 % sensitivity, 68.8 % specificity, 68.9 % F1-score, and 72.6 % area under the receiver operating characteristic curve (AUC). CONCLUSIONS:Extracted quantitative sMRI features supported ASD-versus-typical-control classification.
OBJECTIVES:Parkinson's disease (PD) can affect vocal production before severe motor symptoms become apparent. This study presents a device-oriented dual-route voice-screening framework that combines interpretable acoustic descriptors with spectrogram-based transfer learning for PD assessment support. METHODS:Publicly available sustained-vowel /a/ recordings from the Parkinson Speech Dataset with Multiple Types of Sound Recordings were used for algorithmic evaluation. The proposed handheld device, dashboard, and connectivity workflow were treated as a conceptual deployment scenario. The acoustic route generated an integrated acoustic-feature vector and grouped representation-level summaries of selected descriptor families. The spectrogram route generated RGB color-mapped and grayscale time-frequency representations, which were processed using an EfficientNet-B0 transfer-learning convolutional neural network. RESULTS:Among conventional acoustic-feature classifiers, the Ensemble Boosting Classifier achieved the highest AUC point estimate for the integrated acoustic-feature route. Grouped representation-level evidence showed weak-to-moderate discrimination for acoustic descriptor families. RGB and grayscale spectrograms processed through the EfficientNet-B0 route produced reported AUC point estimates of 0.93 and 0.89, respectively, within the available public-dataset evaluation. CONCLUSIONS:The proposed framework combines interpretable acoustic descriptors and spectrogram-based transfer learning within a device-oriented screening-support workflow. Integrated acoustic features support interpretable reporting, while spectrogram-based transfer learning provides higher representation-level discrimination. Future validation with physical prototypes, standardized acquisition protocols, and independent clinical cohorts is required before clinical deployment.
OBJECTIVES:To evaluate the feasibility and clinical utility of cardiovascular four-dimensional flow magnetic resonance imaging (4D Flow MRI) for multidirectional hemodynamic visualization and quantitative assessment. METHODS:This single-centre observational study included 30 cardiovascular MRI cases acquired at Huaqiao University Affiliated Strait Hospital using a Siemens MAGNETOM Skyra 3.0-T system with an 18-channel phased-array coil. Time-resolved three-dimensional velocity-encoded 4D Flow MRI was performed with retrospective electrocardiographic gating, respiratory compensation, and region-adjusted velocity encoding. Acquisition parameters included TR/TE 38.88/6.88 ms, flip angle 8°, matrix 256 × 256, field of view 300 mm, in-plane spatial resolution 1.4 × 1.4 mm, slice thickness 3 mm, and slice gap 0.6 mm. Preprocessing corrected background phase offsets, Maxwell terms, velocity aliasing, and noise artefacts. RESULTS:4D Flow MRI enabled assessment of intracardiac shunting, pulmonary regurgitation, Fontan circulation, bicuspid aortic valve hemodynamics, aortic regurgitation, and atrioventricular valve dysfunction. Quantified parameters included pulmonary-to-systemic flow ratio, shunt volume, regurgitant fraction, peak velocity, pressure gradient, wall shear stress, vorticity, helicity, kinetic energy, turbulent kinetic energy, oscillatory shear index, and energy loss. CONCLUSIONS:Cardiovascular 4D Flow MRI supports patient-specific hemodynamic assessment across complex cardiovascular conditions within a practical single-centre clinical workflow with retrospective quantification.
OBJECTIVES:Ascending aortic growth in BAV varies widely, and baseline diameter insufficiently predicts individual risk; we assessed whether abnormal regional WSS surface burden relates to later AAo enlargement during longitudinal follow-up. METHODS:The analysis included 37 patients with BAV and 66 healthy control subjects matched for age and sex. Baseline thoracic aortic phase-contrast/4D-flow MRI was acquired using a 3.0-T Siemens MAGNETOM Skyra scanner to evaluate flow-related parameters across the aortic root, AAo, arch, and proximal descending thoracic aorta. The velocity-encoded 3D acquisition was planned to include the thoracic aorta, and axial or oblique axial source images were used for anatomical lo calization and AAo diameter measurement. Imaging parameters included TR 38.88 ms, TE 68.8 ms, flip angle 8°, matrix 256 × 256, field of view 300 mm, spatial resolution 1.4 × 1.4 mm2, slice thickness 3 mm, and interslice gap 0.6 mm. Patient WSS maps were evaluated against healthy-control 95 % confidence interval reference maps, and higher AAo growth was defined as annual enlargement greater than 0.25 mm/year. RESULTS:During 6.2-year median follow-up, higher-growth patients showed greater elevated AAo WSS area (20.6 vs. 6.0 %; p=0.021). CONCLUSIONS:Phase-contrast MRI-derived elevated WSS area may support individualized BAV surveillance.
OBJECTIVES:To develop an artificial intelligence (AI) model to predict cesarean scar diverticulum (CSD) risk following cesarean scar pregnancy (CSP) for early clinical risk stratification. METHODS:A total of 120 CSP patients were retrospectively enrolled and randomly split into training (n=84) and test (n=36) cohorts. Data on clinical, laboratory, and ultrasonographic parameters, such as uterine scar muscle thickness, gestational sac diameter, and cesarean history, were gathered. A deep convolutional neural network (CNN) using a Faster R-CNN framework was trained to predict postoperative CSD, and feature importance analysis identified key predictors. RESULTS:Indicated significant differences in uterine scar muscle thickness, clinical classification, gestational sac diameter, and coagulation parameters between CSD and non-CSD groups (p<0.05). The AI-CNN model achieved an accuracy of 0.944, sensitivity of 0.917, and specificity of 0.958 in the test set. Key predictors included uterine scar muscle thickness of ≤0.2 cm, clinical classification type II-III, and a history of two or more cesarean sections. CONCLUSIONS:The AI-based CNN model provides accurate prediction of CSD risk after CSP. Identified preoperative indicators may guide clinical decision-making and targeted postoperative surveillance.
The integration of artificial intelligence (AI) in digital pathology has shown significant promise in advancing cancer diagnostics, grading, and treatment response prediction. However, widespread development and deployment of robust AI models face critical challenges due to data silos, privacy concerns, and the need for large-scale multi-institutional datasets. Federated Learning (FL) presents a transformative approach by enabling collaborative model training across hospitals without direct data sharing. In this review, we summarize recent developments in FL as applied to digital pathology, highlight pioneering use cases, and explore the technical, regulatory, and ethical hurdles. We discuss how FL can enable scalable, privacy-preserving AI models, and outline future directions for standardizing and validating FL-based approaches in clinical workflows.
OBJECTIVES:To develop and evaluate an automated deep learning framework for scoliosis detection from spinal radiographs and to compare the performance of a single-stage YOLOv11-based approach with a two-stage ROI-guided ResNet model. METHODS:A publicly available spinal X-ray dataset was analyzed for key radiographic features, including spinal alignment, lateral curvature, vertebral asymmetry, and overall spinal contour. Two automated strategies were compared: a YOLOv11-based framework for spinal region detection and direct classification, and a two-stage approach in which the spinal region of interest was first localized using bounding box regression and then classified with ResNet. RESULTS:The YOLOv11-based framework outperformed the ROI-based ResNet approach in distinguishing scoliosis from non-scoliosis radiographs. It captured scoliosis-related deformity patterns more effectively and demonstrated stronger diagnostic stability across representative cases. CONCLUSIONS:The YOLOv11-based framework demonstrated superior performance over the two-stage ResNet strategy for automated scoliosis recognition and may serve as a useful adjunct for radiographic screening.
OBJECTIVES:To evaluate a multimodal deep learning model integrating preoperative transvaginal ultrasound (TVUS)-based radiomics features and clinical indicators for predicting 1-year postoperative recurrence of endometrial polyps (EP) after hysteroscopic polypectomy. METHODS:A total of 116 patients with pathologically confirmed EP were assigned to a training cohort (n=81) and validation cohort (n=35). Radiomics features were extracted from TVUS images, and deep learning features were obtained using ResNet-based networks. These features, with clinical variables, were combined to build a multimodal model. Feature selection in the training cohort used reproducibility filtering (intraclass correlation coefficient [ICC] >0.80), univariate analysis, Pearson correlation (|r|>0.90), and least absolute shrinkage and selection operator (LASSO) regression. Model performance was evaluated by receiver operating characteristic (ROC) curves, area under the curve (AUC), calibration, and decision curve analysis (DCA). RESULTS:The multimodal model achieved AUCs of 0.941 (95 % CI: 0.897-0.985) and 0.922 (95 % CI: 0.852-0.992) in training and validation cohorts, outperforming clinical (0.812, 0.791) and radiomics-only models (0.871, 0.843). DeLong tests were significant (p<0.05). DCA showed higher clinical net benefit. CONCLUSIONS:This multimodal model effectively predicts 1-year recurrence after hysteroscopic EP resection, supporting individualized postoperative management.
BACKGROUND:Radiological analysis of splenic images was conducted to evaluate postoperative hemorrhage following partial splenectomy in 50 patients. Computed tomography (CT) was used as the primary imaging method for postoperative assessment. The study included 41 males and nine females aged 16-79 years (mean age 52) treated between December 2019 and July 2023. METHODS:Clinical management focused on identifying preoperative bleeding risks, monitoring blood pressure, managing anticoagulant therapy, and early detection of postoperative hemorrhage. Patient care also included vital-sign monitoring, psychological support, predictive medical strategies, and the use of modern communication tools for patient education. Patients who developed hemorrhage were treated with either exploratory laparotomy or conservative management based on their clinical condition. RESULTS:The average hospital stay was 12.5 days (range 5-18 days). CT imaging revealed hemorrhagic splenic lesions before and after treatment. Statistical analysis showed a strong correlation (r=0.94) between CT findings and treatment outcomes, confirming the high diagnostic value of CT in detecting splenic hemorrhage and guiding clinical management. CONCLUSIONS:Comprehensive perioperative management combined with predictive medical strategies can reduce recurrence of postoperative complications and improve surgical workflow for traumatic splenic injury.
OBJECTIVES:To develop an MS-Res-AttU-Net-based deep learning framework for automatic measurement of vertebral compression ratio (VCR) on lumbar magnetic resonance images and to evaluate its value for image-based assessment of lumbar vertebral fractures. METHODS:This retrospective study included 92 patients with lumbar vertebral fractures who underwent sagittal T2-weighted MRI. An MS-Res-AttU-Net framework was constructed for vertebral segmentation and automatic VCR calculation. The dataset was divided into a training cohort (n=64) and an independent test cohort (n=28). Segmentation performance was assessed using sensitivity, specificity, accuracy, and Dice similarity coefficient. Agreement between automated and manual VCR measurements was evaluated using correlation, intraclass correlation coefficient, and Bland-Altman analysis. An ablation study was further performed to assess the contribution of residual, attention, and multi-scale refinement modules. RESULTS:The final MS-Res-AttU-Net achieved stable segmentation performance and showed close agreement between automated and manual VCR measurements. The ablation study demonstrated progressive improvement in both segmentation quality and downstream VCR estimation, while A qualitative comparison of four lumbar MR cases showed MS-Res-AttU-Net produced the smoothest and most accurate vertebral contours. CONCLUSIONS:Automatic VCR measurement on lumbar MR images is feasible and clinically interpretable. The MS-Res-AttU-Net-based framework may provide a rapid and objective quantitative tool for lumbar fracture evaluation.
OBJECTIVES:We developed a multimodal fusion model combining clinical data and deep transfer learning for early progressive cerebral contusion (PCC) prediction, providing precise clinical support for treatment decisions. METHODS:Using a single-center retrospective cohort design, we analyzed 196 cerebral contusion patients between January 2022 and June 2024. PCC was characterized by a contusion volume increase of at least 30 % on CT scans within 24 h. Patients were categorized into a progression group (n=98) and a non-progression group (n=98). The dataset was split into a training coh59 participants, maintaining a 7:3 ratort of 137 participants and a validation cohort of io. A nomogram was developed by combining ResNet-50-based deep transfer learning features with clinical variables. Model performance was assessed through ROC curves, calibration plots, and decision curve analysis, while Grad-CAM was used to evaluate interpretability. RESULTS:The integrated nomogram demonstrated superior performance with AUC values of 0.999 (95 % CI: 0.998-1.000) in the training cohort and 0.972 (95 % CI: 0.939-1.000) in the validation cohort, surpassing the standalone DTL and clinical models. Grad-CAM demonstrated accurate lesion localization. CONCLUSIONS:The multimodal fusion model integrating DTL and clinical features shows excellent predictive performance and significant clinical value in early PCC prediction.
OBJECTIVES:Early hematoma expansion is a major determinant of poor outcome in hypertensive basal ganglia hemorrhage. This study evaluated whether CT-based radiomic texture analysis could improve early prediction of hematoma expansion. METHODS:A retrospective cohort of 104 patients with hypertensive basal ganglia hemorrhage who underwent baselinef CT within 6 h of symptom onset and follow-up CT within 48 h was analyzed. Hematoma regions of interest were manually segmented, and 256 texture features were extracted using MaZda. Fisher's score, probability of error and average correlation coefficient, and mutual information were used for dimensionality reduction. Classification performance was assessed using raw data analysis, principal component analysis, linear discriminant analysis, and nonlinear discriminant analysis, followed by ROC analysis. RESULTS:Hematoma expansion occurred in 40 of 104 patients (38.4 %). Nonlinear discriminant analysis showed the lowest misclassification rate overall, including 0 under the POE-ACC feature set. ROC analysis demonstrated good diagnostic performance for several texture features, with S(3, -3)Difvarnc (AUC 0.944), S(4, -4)Difvarnc (AUC 0.942), and GrVariance (AUC 0.917) showing the strongest predictive value. CONCLUSIONS:CT-based texture analysis provides quantitative imaging biomarkers that may support early risk stratification of hematoma expansion in basal ganglia hemorrhage.
BACKGROUND:Existing medical image generation tasks primarily employ Generative Adversarial Networks (GANs), which perform poorly on datasets with temporal characteristics and suffer from slow generation speed and mode collapse. METHODS:In response to this question, this study puts forward a temporal conditional diffusion model based on a dual U-Net structure, which leverages the dual U-Net to extract rich detail information within a denoising diffusion framework while incorporating temporal information as a condition to guide the generation of 4D cardiac datasets with temporal features. Additionally, a deformation field is utilized to accelerate medical image generation. RESULTS:Experimental results show that compared to existing methods, the proposed approach can generate dynamic scan time frames while maintaining strong continuity and temporal consistency in both transverse and longitudinal spatial dimensions. In addition, the synthesized images are highly similar to those captured in reality. The proposed method effectively preserves anatomical structural details, making it highly suitable for medical image generation tasks.
OBJECTIVES:To predict abnormal pulmonary artery hemodynamics caused by ventricular septal defect (VSD) using Physics-Informed Neural Networks (PINN) and address the challenges of high computational cost in traditional Computational Fluid Dynamics (CFD) and difficulty in obtaining measurement data. METHODS:The PINN model was trained using boundary conditions and scattered clinical CFD data, with dynamic weighting factors incorporated to enhance training efficiency and optimize predictions. Model outputs for blood flow velocity and pressure were subsequently evaluated against CFD simulation results. RESULTS:The PINN accurately reproduced velocity and pressure fields across pulmonary artery models using only boundary conditions and sparse internal measurements. For velocity prediction, the average RMSE, MAE, and MRE for components u, v, and w ranged from 0.274 to 0.832 %, 0.448-1.096 %, and 0.833-1.341 %, respectively. For pressure prediction, the average RMSE, MAE, and MRE ranged from 2.953 to 5.145 %, 3.264-5.679 %, and 0.376-0.565 %, respectively. These findings demonstrate that the framework generalizes well and provides reliable hemodynamic estimation with limited input data. CONCLUSIONS:The PINN model compensates for incomplete measurement data through physical constraints, enabling rapid and accurate prediction of pulmonary artery hemodynamics and offering a promising non-invasive alternative for pulmonary artery pressure measurement.
OBJECTIVES:This study aims to propose an algorithm to select feature-rich patches from histopathology images, eliminating irrelevant regions, to support the computer-aided diagnosis of oral cancer. METHODS:To identify interpretable and diverse feature-rich patches, Haralick and local binary pattern (LBP) features are extracted and cosine similarity is calculated between patches. A trained convolutional neural network (CNN) model with spatial attention, along with a SHapley Additive exPlanation (SHAP) explainer, is used to estimate each patch's contribution using SHAP values. These values, combined with cosine similarity, guide the selection of four patches that significantly influence the diagnosis while ensuring diversity. Features from the selected patches are extracted using a ResNet-50 model, concatenated, and normalized using Z-score normalization. The resulting feature vector is used to classify images as normal or cancerous. RESULTS:The approach was tested on a public dataset of 1,224 oral histopathology images from 230 patients, covering 100× and 400× magnifications. It achieved 97.53 % accuracy, 97.68 % precision, and 97.36 % sensitivity using a voting classifier, and 97.93 %, 99.43 %, and 97.76 % respectively with a stacking classifier, outperforming existing cancer detection methods. CONCLUSIONS:The proposed algorithm effectively classifies oral biopsy images with minimal training, enhancing diagnostic reliability and offering valuable support for pathologists.
