BioMedical Engineering OnLine最新文献

Background: Diabetic foot (DF) is a severe complication of type 2 diabetes mellitus (T2DM), contributing to significant morbidity and healthcare costs globally. Early prediction and intervention are critical for preventing amputations and improving patient outcomes. However, traditional statistical methods lack the capacity to handle high-dimensional clinical data and identify optimal predictive features. This study aimed to develop and validate machine learning models for DF risk prediction using feature selection strategies based on binary logistic regression and information theory.

Methods: A retrospective cohort of 1,179 patients (95 DF cases, 1,084 T2DM controls) was analyzed using clinical and biochemical data from 2019 to 2025. Three data sets were constructed: (1) original features; (2) features selected via binary logistic regression (F1); and (3) features selected via information-theoretic global learning (F2). Six models-extreme learning machine (ELM), kernel extreme learning machine (KELM), and their variants trained on the three data sets-were evaluated using fivefold cross-validation. Performance metrics included area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and computational efficiency.

Results: Age, blood-urea-nitrogen (BUN), homocysteine (Hcy), albumin (ALB), and fasting blood glucose (FBG) were identified as independent DF risk factors. The information theory-based KELM (IT-KELM) model achieved the highest AUC of 0.799 (sensitivity: 0.792 and specificity: 0.710) on F2, outperforming other models. Feature selection improved predictive accuracy while reducing computational time, with IT-KELM requiring 0.138 s for training and 0.0023 s for testing. The SHAP summary dot plot and bar chart revealed that the top five features contributing to the model were TP, RBC, ALB, BMI and HB.

Conclusions: Integrating information theory with KELM enhances DF risk prediction by optimizing feature subsets and leveraging nonlinear kernel mapping. The IT-KELM model demonstrates robust diagnostic performance and clinical feasibility for early DF screening. Future multi-center studies are needed to validate generalizability and refine model interpretability in real-world settings. This approach provides a cost-effective tool for precision medicine in diabetes care.

Background: Conventional shape capture in prosthetics and orthotics (P&O) relies on plaster casting of a positive mold, which is then hand-rectified to the desired shape. While effective under expert practice, this workflow is labor-intensive, equipment-dependent, and difficult to archive or share. Digital approaches using structured-light scanners address some of these limitations but remain costly and require dedicated training. This study evaluated whether smartphone photogrammetry can accurately and reliably capture prosthetic and orthotic cast geometries, and assessed its usability in comparison to a clinical structured-light scanner for integration into clinical workflows.

Results: A clinical-grade structured-light scanner (EinScan H2) served as the reference and demonstrated small volumetric and dimensional errors, at 0.21 ± 0.15% and 0.35 ± 0.18 mm, with intraclass correlation coefficients (ICCs) greater than 0.9999. Relative to this reference, across 12 cast models (upper limb, lower limb, and ankle-foot orthosis), smartphone photogrammetry achieved a volumetric error of 0.89 ± 0.68% and a dimensional error of 0.89 ± 0.51 mm; the mean surface point-to-point distance was 0.24 ± 0.19 mm. Reliability across operators was near-perfect (ICCs ≥ 0.9997). Usability data showed approximately 62 photographs and 88 s per capture for photogrammetry (about 12 min cloud processing) versus 34 s capture (about 85 s desktop processing) for the reference scanner. Photogrammetry scored higher on the System Usability Scale (79 versus 58).

Conclusions: On casts, smartphone photogrammetry produced accurate and reliable meshes with favorable perceived usability and minimal hardware demands. These findings support its integration into digital workflows in P&O, particularly for scanning rectified positive casts and other stable geometries. Further multi-site evaluations on live limbs should determine acceptable capture-time thresholds and effective stabilization strategies to ensure clinical feasibility in routine practice.

Individuals with incomplete spinal cord injury (iSCI) often fall due to decreased sensorimotor integration. Functional electrical stimulation (FES) therapy combined with visual feedback balance training (VFBT), termed FES+VFBT, can effectively improve standing balance in iSCI populations. Although promising, the need for force plates (FP), which are expensive and bulky, limits the translation of these methods to clinical and home settings. In this work, we propose a solution by replacing FP with Wii Balance Board (WBB), allowing for more accessible FES+VFBT at a lower cost in both clinical and community settings. Our investigations on ten non-injured participants reveal that WBB-based estimated center of mass (COM) has low prediction error and high correlation in both anteroposterior (RMSE: 4.13 ± 0.69 mm, r: 0.94 ± 0.02) and mediolateral directions (RMSE: 6.25 ± 1.80 mm, r: 0.92 ± 0.04) with ground FP-estimated COM, resulting in similar stimulation patterns obtained with the WBB-based approach, indicating that the WBB-based FES+VFBT system could yield a more accessible therapeutic strategy for balance rehabilitation in iSCI.

Background: Smart patch healthcare devices are emerging as a distinct user interface in decoding the bidirectional interaction of the five sense organs. Powered by recent advancements in nano-materials, and artificial intelligence predictions, smart patches could understand the immune response of the body by analysing the biofluids, microenvironment and analytes in the five sense organs. These eminent potentials in smart patches, inspired the necessity for a review. Thus, this review aims to bring in to the limelight the current progress in smart patch technologies, highlighting their functions, opportunities and challenges in healthcare applications.

Methods: A comprehensive review of literature was conducted focusing on smart patches designed for skin, ocular, cochlear, oral, and nasal applications. Further, the review is structured emphasising details on materials used, fabrication methods adapted, sensing mechanisms employed, enabling technologies such as artificial intelligence and Internet of Things.

Results: The review analysis revealed that smart patches play a multifaceted role in healthcare applications providing (i) continuous health monitoring, (ii) controlled drug delivery, (iii) supports tissue regeneration and (iv) enables modulation of nerve responses. Further, smart patch integration with Internet of Things (IoT) capabilities enables remote healthcare solutions which benefits both physician and patient communities equally. Despite these progresses, challenges remain in term of biocompatibility of the materials chosen, long-term use and stability of the patch, data security and large-scale manufacturing.

Conclusion: Smart patches hold transformative potential in biomedical engineering by bridging biosensing, therapeutic, and digital healthcare domains. This article provides an in-depth review of the current advancements, identifying the existing challenges and emerging opportunities in the field of smart patch research, and thus could guide future research and development. With its broad scope, this review would act as a valuable resource for both researchers and healthcare innovators working towards next-generation biomedical devices.

Purpose: To develop a cloud-based automated treatment planning system for intensity-modulated radiation therapy and evaluate its efficacy and safety for tumors in various anatomical sites under general clinical scenarios.

Results: All the plans from both groups satisfy the PTV prescription dose coverage requirement of at least 95% of the PTV volume. The mean HI of plan A group and plan B group is 0.084 and 0.081, respectively, with no statistically significant difference from those of plan C group. The mean CI, PQM, OOT and POT are 0.806, 77.55. 410 s and 185 s for plan A group, and 0.841, 76.87, 515.1 s and 271.1 s for plan B group, which were significantly superior than those of plan C group except for the CI of plan A group. There is no statistically significant difference between the dose accuracies of plan B and plan C groups.

Conclusions: It is concluded that the overall efficacy and safety of the Desargues Cloud TPS are not significantly different to those of Varian Eclipse, while some efficacy indicators of plans generated from automatic planning without or with manual adjustments are even significantly superior to those of fully manual plans from Eclipse. The cloud-based automatic treatment planning additionally increase the efficiency of treatment planning process and facilitate the sharing of planning knowledge.

Materials and methods: The cloud-based automatic radiation treatment planning system, Desargues Cloud TPS, was designed and developed based on browser/server mode, where all the computing intensive functions were deployed on the server and user interfaces were implemented on the web. The communication between the browser and the server was through the local area network (LAN) of a radiotherapy institution. The automatic treatment planning module adopted a hybrid of both knowledge-based planning (KBP) and protocol-based automatic iterative optimization (PB-AIO), consisting of three steps: beam angle optimization (BAO), beam fluence optimization (BFO) and machine parameter optimization (MPO). 53 patients from two institutions have been enrolled in a multi-center self-controlled clinical validation. For each patient, three IMRT plans were designed. The plan A and B were designed on Desargues Cloud TPS using automatic planning without and with manual adjustments, respectively. The plan C was designed on Varian Eclipse TPS using fully manual planning. The efficacy indicators were heterogeneous index, conformity index, plan quality metric, overall operation time and plan optimization time. The safety indicators were gamma indices of dose verification.

Objective: This study aimed to develop and validate a high-precision brain age prediction model by integrating multimodal MRI radiomics features from T1- and T2-weighted images with deep learning. The model was trained on healthy individuals for chronological age estimation and applied to patients with insomnia to calculate the Brain Age Gap (BAG), evaluating whether chronic insomnia is associated with accelerated brain aging.

Methods: A total of 1,200 participants were retrospectively included, comprising 942 healthy controls and 258 patients with insomnia. Healthy data were obtained from the IXI public dataset and Shenzhen Hospital (Futian), Guangzhou University of Chinese Medicine. All insomnia patients were recruited from the same hospital. T1- and T2-weighted MRI underwent standardized preprocessing, including resampling, gray-level discretization, and automated segmentation for radiomics feature extraction. After variance-based feature selection, multimodal features were combined to construct a deep learning regression model trained on healthy subjects and evaluated using mean absolute error (MAE), root mean square error (RMSE), and R². The model was then applied to the insomnia cohort to estimate BAG, followed by age-bias correction and group comparisons.

Results: Three models were constructed: T1-based, T2-based, and multimodal fusion. In validation, the T1 model achieved MAE of 7.58 years (R² = 0.57), the T2 model 7.90 years (R² = 0.51), and the fusion model 6.42 years (R² = 0.68; all p < 0.001). The insomnia group showed significantly higher BAG than controls both before (8.10 ± 8.57 vs. 1.26 ± 8.30 years, p < 0.001) and after age correction (1.60 ± 6.49 vs. - 2.18 ± 7.75 years, p < 0.001).

Conclusion: The multimodal MRI radiomics-deep learning fusion model enables accurate brain age prediction and reveals evidence of accelerated brain aging in patients with insomnia.

Background: Electrical impedance tomography (EIT) is a non-invasive bedside tool for real-time regional ventilation data, but its correlation with outcomes in different critical illnesses remains unclear.

Methods: This retrospective study included 108 ICU patients (liver disease, n = 48; respiratory failure, n = 36; gastrointestinal bleeding, n = 24) who underwent EIT monitoring between December 2023 and March 2025. EIT measured tidal volume percentage in four regions of interest (ROIs), with 28-day mortality as the primary outcome. Multivariate logistic regression adjusted for key confounders was used for analysis.

Results: Ventilation distribution patterns differed significantly among the three disease groups. The liver disease group showed predominant ventilation in ROI 1-2 (anterior regions), with a mean dorsal-to-ventral ratio (DVR) of 0.41 ± 0.18. The respiratory failure group exhibited more homogeneous distribution with a DVR of 0.76 ± 0.34. Patients with higher ventilation heterogeneity (coefficient of variation > 40%) had significantly higher 28-day mortality (29.5% vs 16.7%, p = 0.022), longer duration of mechanical ventilation (8.7 vs 5.4 days, p = 0.012), and fewer ventilator-free days (14.3 vs 19.6 days, p = 0.009). Multivariate analysis identified DVR < 0.4 (OR 2.84, 95% CI 1.46-5.53, p = 0.002) and ventilation heterogeneity (OR 2.31, 95% CI 1.18-4.52, p = 0.014) as independent predictors of 28-day mortality after adjusting for disease severity, age, and mechanical ventilation parameters.

Conclusions: Regional ventilation patterns vary by underlying disease. Greater heterogeneity and lower DVR independently correlate with worse outcomes. EIT-derived parameters may be prognostic indicators and therapeutic targets for optimizing ventilation in critical illness.

Objective: To provide a critical and clinically oriented synthesis of recent deep learning developments for breast cancer imaging across major modalities, with emphasis on model architectures, dataset characteristics, methodological quality, and implications for clinical translation.

Methods: Following PRISMA guidelines, we systematically searched PubMed, Scopus, Web of Science, ScienceDirect, and Google Scholar for studies published from 2020 to 2024 on deep learning applied to breast imaging. Sixty-five studies using convolutional neural networks (CNNs), Transformers, or hybrid architectures were included. Datasets were comparatively profiled, and study quality and risk of bias were appraised using QUADAS-2.

Results: CNN-based classifiers, particularly on mammography and pathology, commonly achieved median accuracies above 90% and AUCs around or above 0.95, while CNN detectors reported high sensitivities and mid-90% accuracies, supporting their potential role as second readers. CNN-derived U-Net variants dominated segmentation tasks, yielding high Dice and IoU values for tumour and fibroglandular-tissue delineation. Transformer and hybrid models showed advantages when global context, multi-view inputs or volumetric data were critical (e.g. dense breasts, DBT, DCE-MRI), where they improved lesion localisation and patient-level risk stratification. However, QUADAS-2 and dataset profiling revealed substantial limitations: most studies were retrospective, single-centre and class-imbalanced, with narrow demographic representation, heterogeneous reference standards and scarce external or prospective validation. These factors raise concerns about bias, overfitting, fairness and robustness in real-world deployment. Only a minority of studies systematically addressed interpretability, workflow integration or regulatory requirements.

Conclusions: Deep learning offers considerable promise to support early detection, risk stratification and workflow efficiency across breast imaging modalities, with CNNs and Transformers providing complementary strengths for local fine-detail versus global contextual modelling. Nevertheless, the current evidence base is constrained by heterogeneous designs, limited reporting of study quality and biased datasets, so reported performance should not be interpreted as definitive proof of clinical readiness. Future research should prioritise multi-centre, demographically diverse cohorts, transparent quality assessment, external and prospective validation, and evaluation of reader and workflow impact. Developing explainable, fairness-aware and privacy-preserving systems-such as those enabled by interpretable architectures and federated learning-will be essential for safe and equitable translation of deep learning tools into routine breast cancer care.

Background: Wearable cameras provide a means to assess hand function in individuals with spinal cord injury (SCI) beyond clinical settings. Previous studies have found that clinicians acknowledge the potential of egocentric video to monitor and inform rehabilitation. Nonetheless, the need for time-intensive manual review of the footage remains a challenge to its integration into clinical practice. To address this barrier, we investigated the utility of video summarization for egocentric videos of hand use after SCI.

Methods: A dataset comprising 316 egocentric videos from 20 individuals with cervical SCI was used. Individuals wore head-mounted cameras to record daily activities in their home. Three unsupervised video summarization algorithms were applied: DR-DSN (reinforcement learning), CTVSUM (contrastive learning), and CA-SUM (attention-based learning). The resulting summaries were manually evaluated on a subset of five videos (each summarized by all three algorithms) by 15 participants using five criteria rated on a 5-point Likert scale: (C1) inclusion of hand movements, (C2) visibility of difficulties and compensation, (C3) contextual clarity, (C4) depiction of hand function, and (C5) preservation of key information. Additionally, summaries were assessed using computational metrics: coverage, temporal distribution, diversity, and representativeness.

Results: An average manual rating of 3.7 ± 1.2 was observed. Ratings differed significantly across both evaluation criteria (F = 13.69, p < 0.001, ${\eta }^2$  = 0.167) and algorithms (F = 24.00, p < 0.001, ${\eta }^2$  = 0.103). In particular, summaries were rated higher for C3 and lower for C2, while CA-SUM consistently received the highest scores. Among the computational metrics, diversity showed a strong negative association with manual ratings (b = -4.8, p = 0.032, R² = 0.827), while representativeness was positively associated (b = 17.8, p = 0.047, R² = 0.779).

Conclusion: All three algorithms produced adequate video summaries that captured essential content. However, enhancing the depiction of aspects such as functional difficulties and compensatory strategies could further improve the clinical value of the summaries. Moreover, discrepancies between computational and manual evaluations highlight the need to train algorithms on more human-centered criteria. Overall, this work demonstrates the potential of automatic video summarization to support the integration of wearable cameras into outpatient SCI rehabilitation.