Computers in biology and medicine最新文献

Introduction: Adult spinal deformity (ASD) involves complex three-dimensional (3D) spinal malalignments that impair mobility and stability. Current clinical assessments rely on static, two-dimensional (2D) radiographs, which fail to capture the 3D dynamics essential for comprehensive evaluations. While musculoskeletal models with marker-based motion analysis offer insights into kinematics, generic models fail to replicate the 3D deformities in ASD. This study introduces an automated workflow to generate image-based subject-specific models, capturing individual spinal geometry and alignment to enable analysis of 3D dynamics in patients with ASD.

Methods: A retrospective dataset of 13 deformity subjects was used to develop and evaluate the workflow. Spinopelvic bones were automatically segmented, followed by spinal joint and alignment definition. The accuracy of 3D spinal alignment was validated by simulating upright standing and bending positions as captured with biplanar radiography. 3D position and rotation differences were calculated against biplanar imaging-based reference markers.

Results: 3D position differences across spinal markers averaged 2.2 ± 1.6 mm in the upright, and 3.0 ± 1.9 mm in the bending poses. In bending simulations, differences were comparable to Overbergh et al. (2020) who achieved mean errors 3.0 ± 2.0 mm. 3D rotation differences averaged 3.5 ± 1.7° in the upright, and 5.3 ± 2.6° in the bending poses. The rotation differences in bending compared well with the method of Overbergh et al. (2020) being 5.1 ± 3.0° on average.

Discussion: The proposed workflow enabled creation of image-based subject-specific models of patients with ASD, with anatomically correct spinopelvic bone geometries, intervertebral joints, and 3D alignment.

Integrating Artificial Intelligence (AI) into Survival Analysis (SA) has advanced predictive modeling in healthcare, enabling precise and personalized predictions of time-to-event outcomes, such as patient survival. However, real-world SA datasets often suffer from data scarcity, heterogeneity, and privacy constraints, which limit the applicability of traditional and modern AI methods. To address these challenges, we propose the Federated Synthetic Data Sharing (FedSDS) framework, which integrates synthetic data generation with Federated Learning (FL). For SA, we leverage SAVAE, a state-of-the-art model for complex datasets. Using the Variational Autoencoder-Bayesian Gaussian Mixture model enhanced with artificial inductive bias, FedSDS generates high-quality synthetic data locally and shares them among nodes, enabling collaborative model training without direct data sharing. FedSDS introduces a biased aggregation strategy that aligns synthetic data with local distributions, outperforming traditional FL methods, such as Federated Average. Validated under independent and identically distributed (IID) and non-IID scenarios, FedSDS mitigates data imbalances and heterogeneity, showing significant performance improvements in scarce and heterogeneous data. The proposed framework offers a scalable and privacy-preserving solution for SA in decentralized environments. By enhancing model generalizability and robustness, FedSDS provides a promising path forward for collaborative analytics in healthcare, paving the way for improved patient outcomes and greater adoption of federated techniques in real-world applications.

Digital therapeutics, enabled by advanced machine learning algorithms and medical wearable devices, offer a promising approach to streamline diagnostics and improve access to healthcare. Within this framework, automatic sleep scoring can provide accurate and efficient sleep analysis from electrophysiological signals recorded with wearable sensors, such as electroencephalography (EEG). However, the optimal configuration and temporal dynamics of automatic sleep scoring systems remain unclear, especially concerning their performance across different population samples. This study systematically investigates the impact of: (1) electrode setup, (2) temporal scope, and (3) population characteristics on the performance of automatic sleep scoring algorithms. Utilizing a convolutional neural network (CNN), we analyzed various electrode configurations and temporal dynamics using datasets comprising both healthy participants and individuals with sleep apnea. Our findings reveal that sleep scoring based on a single frontal EEG channel demonstrates reliable congruency with human expert scorers, with minimal improvement observed with additional sensors. Moreover, we demonstrate that real-time sleep scoring can be achieved with comparable accuracy to offline methods, which rely on past and future information to classify a window of interest. Remarkably, a notable reduction in decoding accuracy is observed for individuals with sleep apnea compared to healthy participants, highlighting the challenges inherent in accurately assessing sleep stages in clinical populations. Digital solutions for automatic sleep scoring hold promise for facilitating timely diagnoses and personalized treatment plans, with applications extending beyond sleep analysis to include closed-loop neurostimulation interventions. Our findings provide valuable insights into the complexities of automatic sleep scoring and offer considerations for the development of effective and efficient sleep assessment tools in both clinical and research settings.

This study reports the design, synthesis, and comprehensive evaluation of a pyridine-based reduced Schiff base, 2-methoxy-5-methyl-N-(pyridin-2-ylmethyl) aniline (RSB), for its anticancer activity. The molecular structure was confirmed by spectroscopic techniques and single-crystal X-ray diffraction (CCDC: 2419398). Significant hydrogen bonding and π-interactions stabilising the crystal packing were found by Single Crystal XRD and Hirshfeld surface analysis. Compared with cisplatin (IC₅₀ = 27.28 μg/mL), RSB exhibited moderate dose-dependent cytotoxicity with an IC₅₀ of 161.27 μg/mL against the A549 lung carcinoma cell line. Molecular dynamics simulations verified stable and compact binding conformations, which supported the strong binding affinities of RSB toward 1MRV proteins (-7.1 kcal/mol) found in in-silico docking studies. A HOMO-LUMO gap of 6.70 eV was found by DFT analysis, demonstrating the electronic stability of the molecule. All of these results point to the potential of pyridine-based reduced Schiff bases as anticancer agents, indicating the need for additional research and optimisation.

Effective delivery of therapeutic agents to the central nervous system is severely limited by rapid cerebrospinal fluid (CSF) clearance and lack of spatial targeting, hindering treatment of neurodegenerative diseases. Electromagnetic control of nanoparticle-enhanced CSF presents a non-invasive solution, yet fundamental transport mechanisms remain uncharacterized. This work establishes a unified theoretical model for electromagnetically-actuated transport of Casson hybrid nano-CSF containing gold and maghemite nanoparticles in bio-reactor channels with realistic thermal-hydrodynamic boundary conditions, toward optimizing magnetophoretic neurotherapeutic delivery. Coupled momentum-energy equations incorporating Casson rheology, nanoparticle-enhanced properties (Maxwell-Garnett theory), electromagnetic forcing, thermal radiation (Rosseland approximation), porous resistance (Darcy model), and non-uniform heating are solved analytically via Laplace transforms, providing closed-form velocity and temperature solutions. Electromagnetic actuation increases flow velocity by 42% with maghemite nanoparticles showing superior magnetophoretic response; Casson yield stress suppresses backflow, enhancing forward transport by 27%; thermal radiation strengthens convective mixing while asymmetric heating enables targeted deposition; heat sinks reduce wall heat transfer by 35%, limiting thermal drug release. These quantitative insights provide design criteria for magnetically-guided neurotherapeutic systems capable of overcoming biological clearance barriers.

Introduction: Cardiovascular diseases (CVD) are the leading cause of death in developing countries, imposing a significant burden on society. Early detection of patients at higher risk of CVD events could reduce mortality. None of the models currently used for this purpose incorporates insulin resistance (IR), which can be measured using triglyceride and glucose levels. This study aims to explore the effectiveness of the triglyceride-glucose (TyG) index in predicting CVD events using machine learning models.

Methods and materials: This study utilized data from the Mashhad Stroke and Heart Atherosclerotic Disorder (MASHAD) cohort. Patients were evaluated at baseline and monitored for over ten years for CVD events. Eleven machine learning models, including a multilayer perceptron (MLP) and a decision tree, were used to evaluate the predictive value of the TyG index in conjunction with traditional risk factors.

Results: The study population had a CVD event prevalence rate of 10.9%. The average age was 48.08 ± 8.26 years, with 60.0% of participants being female. The mean TyG index was 8.59 ± 0.66. The MLP and AdaBoost classifier models demonstrated the highest predictive accuracy with ROC-AUC scores of 0.77 and 0.766, respectively. The TyG index was identified as the fourth most significant predictor in the AdaBoost Classifier and MLP models, although it ranked lower in other models.

Conclusion: This study highlights the potential benefits of incorporating the TyG index into traditional CVD risk prediction models to enhance accuracy and applicability, especially in developing countries.

Prediction of the 3-D structure of a protein, which can be used to determine its function, is one of the most challenging problems in the bioinformatics field. Protein secondary structure prediction (PSSP) is a crucial step for protein 3-D structure prediction. Recent studies that focused on deep learning methods obtained significant improvements in PSSP. In this study, a novel deep learning model called GraphUnet-SS, which is based on the U-Net architecture and employs convolutional neural networks, graph convolutional networks, and bidirectional long short-term memories, is proposed. The feature set of the model consists of PSI-BLAST position-specific scoring matrices (PSSMs), HHBlits profiles, physico-chemical properties of amino acids, structural profiles for protein secondary structure, and a NoSeq label. A graph is generated using contact map prediction to represent the interactions between amino acids, which is used as input to graph convolutional network layers. The hyperparameters of GraphUnet-SS were optimized using the Bayesian optimization technique. Experimental results show that GraphUnet-SS outperforms the existing methods, and using all layers with depth four is the most suitable version. The source codes of the proposed method are available at https://github.com/ysngrmz/graph_unet_ss and the stand-alone version can be accessed at http://psp.agu.edu.tr/∼psp.

Prostate cancer is a significant global health concern. In the past decade, advances in mathematical models of the disease have led to clinical trials for new therapeutic protocols. However, much of the structure of the prostate and its constituent cell types remains unclear: observations of phenotypic switching between basal and luminal cells, as well as intermediate cell types between them, make classifying the components of the prostate an ongoing challenge. Importantly, a tumour's cell of origin seems to play a role in the aggressiveness and genetic heterogeneity of the cancer. Here we represent the emergence of a tumour by a fit mutant arising amongst healthy, wild-type cells, and we measure the cancer emergence probability and tumour latency by stochastically simulating the system. Taking a compartment model approach, we label cells based on their phenotype and allow for some phenotypic switching between them. We use this to investigate how different structures between cell types impact the emergence and treatment of prostate cancer. We posit that under-regulation of the luminal compartment allows for the explosive population growth associated with cancer. Also, we find that tumours arising in the basal compartment require more time to spread but are more strongly embedded within the prostate, explaining the longer latency and higher aggressiveness and persistence associated with these tumours. Interestingly, the inclusion of a hybrid compartment does not qualitatively change the observations, raising questions as to what mechanistic role intermediate phenotypes play in prostate cancer emergence, if any. Our model contributes to dissecting the relationship between prostate structure and outcomes for cancers arising therein.

Background and objective: Breast cancer is the most common cancer diagnosed among women worldwide, and early detection improves survival outcomes. While medical imaging offers complementary diagnostic information through mammography, ultrasound, and thermal imaging, existing multi-modal computer-aided diagnosis systems suffer from several critical limitations: (a) rigid architectural assumptions which require simultaneous availability of all imaging modalities, (b) entangled feature representations that conflate disease- specific patterns with modality-specific artifacts, and (c) an inability to exploit unpaired multimodal datasets in which patient-level correspondence is unavailable.

Methodology: This work proposes a Hierarchical Contrastive Disentanglement Architecture (HCDA) to address the challenges of multi-modal fusion for breast cancer detection. The proposed work includes: (a) a flexible multi-modal design to adapt variable availability of imaging modalities, (b) a Swin transformer which includes three encoders and orthogonality constraints for explicit hierarchical feature extraction and disentanglement, (c) within-modality contrastive learning to capture semantic representations from unpaired datasets, (d) conditional weighted voting for multimodal feature fusion, and (e) a two-phase sequential training strategy, where disentangled representations are learned in the first phase and fine-tuned for classification in the second. This work is evaluated on total 14,500 images collected from eight publicly available datasets of three modalities (a) mammograms (DDSM, MIAS, INbreast), (b) ultrasound (BrEaST, BUSI, Thammasat, HMSS, and (c) thermal (DMR-IR).

Results: Proposed HCDA achieves strong single-modality performance with mammograms attaining highest accuracy (86.0%) and AUC(0.938), followed by ultrasound (84.9% accuracy, 0.914 AUC), and thermal imaging (80.9% accuracy, 0.954 AUC). For multimodal settings, the model achieves its best dual-modality performance with mammography + ultrasound (84.72% accuracy, 0.926 AUC), while triple-modality fusion yields 80.94% accuracy and a 0.937 AUC.

Conclusions: Evaluation reveals that multi-modal fusion without patient-level correspondence underperform the best single modality (86.0% vs. 84.7%), demonstrating that data structure fundamentally constrains fusion benefits. However, flexible architecture establishes HCDA as a foundation for future multi-modal medical AI research.