Biomedical Physics & Engineering Express最新文献

The enhancement of performance in medical image diagnosis relies on the collaborative representation of features across multiple scales and the ability to accurately capture local lesion characteristics and spatial context. Existing research has shown that conventional convolutional neural networks are constrained by their fixed local receptive field size, which limits their capacity to effectively model global semantic relationships across diverse regions. Although transformers utilizing self-attention mechanisms can capture long-range contextual information, they face challenges in identifying small lesions. To address these issues, this paper presents Hires-Diagnoser, a dual-stream framework for medical image diagnosis that accommodates multiple resolution levels. This framework features a parallel architecture that integrates ConvNeXt and Swin-Transformer branches. The ConvNeXt branch focuses on extracting local texture features through convolutional operations, while the Swin-Transformer branch is responsible for capturing global contextual dependencies via window-based self-attention. Additionally, a cross-modal correlation module (LCA) is introduced to facilitate dynamic interaction and adaptive fusion of features across varying resolutions. Experimental evaluations were conducted on four distinct datasets: RaabinWBC, Brain Tumor MRI, LC25000, and OCT-C8, yielding accuracy rates of 99.45%, 98.01%, 100%, and 97.58%, respectively, thus outperforming existing methods. By leveraging a cross-modal feature interaction mechanism, this framework achieves high performance and meticulous pathological interpretations, providing an effective and highly adaptable solution in the field of medical image diagnosis with significant application potential. .

Purpose.Radiodynamic Therapy (RDT) is an emerging technique that enhances the therapeutic effects of radiation by using photosensitizers to amplify tumor cell damage while minimizing harm to normal tissues. Thisin vitroinvestigation compares the biocompatibility and sensitizing efficacy of two candidate photosensitizers, 5-aminolevulinic acid (5-ALA) and acridine orange (AO), in human breast adenocarcinoma (MCF7) and prostate adenocarcinoma (PC3) cell lines.Materials and Methods.MCF7 and PC3 cell lines were cultured and exposed to a range of 5-ALA and AO concentrations to assess biocompatibility using PrestoBlue viability assays. Based on these results, optimal concentrations were selected for irradiation experiments. Cells were then seeded in T-25 flasks and incubated with 5-ALA or AO prior to receiving 2 Gy or 4 Gy of megavoltage photon radiation (18 MV or 45 MV). Clonogenic assays were performed to determine the surviving fractions of the cells.Results. 5-ALA exhibited a broader biocompatibility profile than AO, remaining non-cytotoxic up to 100 μg ml^-1. In contrast, AO showed cytotoxic effects above 1 μg ml^-1. At 18 MV, limited radiosensitization was observed, except at higher 5-ALA concentrations. However, at 45 MV, both sensitizers significantly reduced cell survival, particularly at 4 Gy. The most pronounced effect was observed with 100 μg ml^-15-ALA, which consistently resulted in lower surviving fractions than AO across both cell lines. Each sensitizer demonstrated differing effectiveness depending on the cell line and photon energy used.Conclusions. Both 5-ALA and AO enhanced the cytotoxic effects of radiation, but 5-ALA demonstrated superior biocompatibility and more consistent radiosensitization across both cell lines. Notably, the effectiveness of both sensitizers increased with higher photon energy, reinforcing the importance of beam energy in RDT design. These results underscore the advantages of 5-ALA over AO and highlight the need to optimize both sensitizer selection and radiation energy in clinical applications.

To enable efficient and accurate retinal lesion segmentation on resource-constrained point-of-care Optical Coherence Tomography (OCT) systems, we propose OCTSeg-UNeXt, an ultralight hybrid Convolution-Multilayer Perceptron (Conv-MLP) network optimized for OCT image analysis. Built upon the UNeXt architecture, our model integrates a Depthwise-Augmented Scale Context (DASC) module for adaptive multi-scale feature aggregation, and a Group Fusion Bridge (GFB) to enhance information interaction between the encoder and decoder. Additionally, we employ a deep supervision strategy during training to improve structural learning and accelerate convergence. We evaluated our model using three publicly available OCT datasets. The results of the comparative experiments and ablation experiments show that our method achieves powerful performance in multiple key indicators. Importantly, our method achieves this high performance with only 0.187 million parameters (Params) and 0.053 G Floating-Point Operations Per second (FLOPs), which is significantly lower than UNeXt (0.246M, 0.086G) and UNet (17M, 30.8G). These findings demonstrate the proposed method's strong potential for deployment in Point-of-Care Imaging (POCI) systems, where computational efficiency and model compactness are crucial.

Polyps are the main biomarkers for diagnosing colorectal cancer. Their early detection and accurate characterization during colonoscopy procedures rely on expert observations. Nevertheless, such a task is prone to errors, particularly in morphological characterization. This work proposes a multi-task representation capable of segmenting polyps and stratifying their malignancy from individual colonoscopy frames. The approach employs a deep representation based on multi-head cross-attention, refined with morphological characterization learned from independent maps according to the degree of polyp malignancy. The proposed method was validated on the BKAI-IGH dataset, comprising 1200 samples (1000 white-light imaging and 200 NICE samples) with fine-grained segmentation masks. The results show an average IoU of 83.5% and a recall of 94%. Additionally, external dataset validation demonstrated the model's generalization capability. Inspired by conventional expert characterization, the proposed method integrates textural and morphological observations, allowing both tasks, polyp segmentation and the corresponding malignancy stratification. The proposed strategy achieves the state-of-the-art performance in public datasets, showing promising results and demonstrating its ability to generate a polyp representation suitable for multiple tasks.

Photoplethysmography (PPG) is widely used in wearable health monitors for tracking fundamental physiological parameters (e.g., heart rate and blood oxygen saturation) and advancing applications requiring high-quality signals-such as blood pressure assessment and cardiac arrhythmia detection. However, motion artifacts and environmental noise significantly degrade the accuracy of PPG-derived physiological measurements, potentially causing false alarms or delayed diagnoses in longitudinal monitoring cohorts. While signal quality assessment (SQA) provides an effective solution, existing methods show insufficient robustness in ambulatory scenarios. This study concentrates on PPG signal quality detection and proposes a robust SQA algorithm for wearable devices under unrestricted daily activities. PPG and acceleration signals were acquired from 54 participants using a self-made physiological monitoring headband during daily activities, segmented into 35712 non-overlapping 5-second epochs. Each epoch was annotated with: (1) PPG signal quality levels (good: 10817; moderate: 14788; poor: 10107), and (2) activity states classified as sedentary, light, moderate, or vigorous-intensity. The dataset was stratified into training (80%) and testing (20%) subsets to maintain proportional representation. Fourteen discriminative features were extracted from four domains: morphological characteristics, time-frequency distributions, physiological parameters estimation consistency and accuracy, and statistical properties of signal dynamics. Four machine learning algorithms were employed to train models for SQA. The random forest (95.6%) achieved the highest accuracy on the test set, but no significant differences (p = 0.471) compared to support vector machine (95.4%), naive Bayes (94.1%), and BP neural network (95.1%). Additionally, the classification accuracy showed no statistically significant variations (p = 0.648) across light (95.3%), moderate (96.0%), and vigorous activity (100%) when compared to sedentary (95.8%). All features exhibited significant differences (p < 0.05) across high/moderate/poor quality segments in all pairwise comparisons.The results indicate that the proposed feature set achieves robust SQA, maintaining consistently high classification accuracy across all activity intensities. This performance stability enables real-time implementation in wearable devices.

Optical coherence tomography (OCT), a non-invasive imaging modality, holds significant clinical value in cardiology and ophthalmology. However, its imaging quality is often constrained by inherently limited resolution, thereby affecting diagnostic utility. For OCT-based diagnosis, enhancing perceptual quality that emphasizes human visual recognition ability and diagnostic effectiveness is crucial. Existing super-resolution methods prioritize reconstruction accuracy (e.g., PSNR optimization) but neglect perceptual quality. To address this, we propose a Multi-level Local-Global feature Fusion Generative Adversarial Network (MLGF-GAN) that systematically integrates local details, global contextual information, and multilevel features to fully exploit the recoverable information in the image. The Local Feature Extractor (LFE) employs Coordinate Attention-enhanced convolutional neural network (CNN) for lesion-focused local feature refinement, and the Global Feature Extractor (GFE) employs shifted-window Transformers to model long-range dependencies. The Multi-level Feature Fusion Structure (MFFS) hierarchically aggregates image features and adaptively processes information at different scales. The multi-scale (×2, ×4, ×8) evaluations conducted on coronary and retinal OCT datasets demonstrate that the proposed model achieves highly competitive perceptual quality across all scales while maintaining reconstruction accuracy. The generated OCT super-resolution images exhibit superior texture detail restoration and spectral consistency, contributing to improved accuracy and reliability in clinical assessment. Furthermore, cross-pathology experiments further demonstrate that the proposed model possesses excellent generalization capability.

Objective.This study presents a fully unsupervised and label-independent radiomic pipeline designed to group different types of ischemic stroke lesions using multimodal Magnetic Resonance Imaging (MRI) . The aim is to address lesion heterogeneity and the absence of annotated outcomes, particularly in settings with limited resources.Approach. Three MRI sequences were analyzed: Fluid Attenuated Inversion Recovery (FLAIR), Apparent Diffusion Coefficient (ADC), and Susceptibility Weighted Imaging (SWI). Lesion identification was performed using percentile-based thresholds, and feature selection was guided by variance filtering with a minimum threshold of 0.001. The complexity of the data was reduced using Uniform Manifold Approximation and Projection (UMAP). Grouping of the lesions was conducted using K-means++, agglomerative hierarchical clustering with Ward linkage, and spectral clustering with a nearest neighbour affinity matrix. The quality and stability of the identified clusters were rigorously evaluated using established internal validation metrics. Feature significance was determined using Kruskal-Wallis testing with Bonferroni correction.Main results. The combination of UMAP with Agglomerative clustering produced the highest silhouette scores of 0.784 for three clusters and 0.778 for five clusters. Consensus stability was optimal, with a Proportion of Ambiguous Clustering score (PAC) of 0.000. Kruskal-Wallis analysis identified 25 significant features for the three-cluster solution and 36 for the five-cluster solution. The most discriminative features originated from ADC and SWI sequences. The five-cluster model revealed finer phenotypic separation and identified five borderline cases with low silhouette coefficients, indicating transitional lesion patterns.Significance. This unsupervised framework enables biologically meaningful lesion stratification without reliance on manual segmentation or outcome labels. It offers a scalable solution for deployment in low-resource environments and provides a robust foundation for future diagnostic and prognostic modelling in stroke imaging.

Major Depressive Disorder (MDD) diagnosis through Electroencephalography (EEG) is hindered by the non-stationary characteristics of neural oscillations and the limited adaptability of conventional classification frameworks. Static ensemble models, which rely on predetermined weight assignments, exhibit suboptimal performance in handling EEG variability induced by inter-individual neurophysiological diversity or environmental artifacts. Meanwhile, monolithic deep learning architectures often suffer from inadequate generalizability in clinical practice. To overcome these limitations, we present an Adaptive Agent-Based Ensemble Learning (AABEL) framework that integrates reinforcement learning (RL) with neurocomputational principles. AABEL pioneers three methodological advancements: (1) RL-Driven Adaptive Weighting: A meta-controller dynamically adjusts the contributions of convolutional (CNN), recurrent (GRU), and attention-based (Transformer) submodels through task-oriented reward signals, resolving the inflexibility of static ensemble paradigms. (2) Multiscale Neurodynamic Feature Fusion: Parallel processing branches extract complementary representations of EEG signals, including spatial-spectral patterns (CNN), temporal-contextual dynamics (GRU), and global interdependencies (Transformer), enabling holistic modeling of neuropathological signatures. (3) End-to-End Reward Propagation: An automated optimization pipeline eliminates manual aggregation rules by directly linking reward calculations to model weight updates. Utilizing the OpenNeuro ds003478 dataset, AABEL achieves superior classification metrics (accuracy: 98.06%, F1-score: 98.20%), outperforming static ensembles (e.g., Fuzzy Ensemble by 96% accuracy). The RL reward mechanism significantly enhances noise robustness, improving classification stability by 3.6%. By integrating dynamic reward-augmented learning with neurosignal processing, AABEL establishes a new paradigm for adaptive EEG-MDD diagnostics. This work bridges computational neuroscience and translational neuroengineering, offering a scalable framework for personalized mental health monitoring.

To segment complex vascular topologies in Optical Coherence Tomography Angiography (OCTA), we introduce DDU-Net. This work addresses the theoretical limitations of standard Swin Transformers, whose internal Multi-Layer Perceptron (MLP) blocks use fixed activation functions, restricting their capacity to model non-linear vascular geometries. We propose the KAN-Swin Transformer, an encoder block that replaces this rigid component with an adaptive operator based on Kolmogorov-Arnold Networks (KANs). This new layer features B-spline-based learnable activation functions on network edges, rather than fixed functions on nodes, empowering the encoder to learn geometrically-aware representations specific to intricate morphologies like bifurcations and high-tortuosity segments. The decoder features a novel dual-path Double Dynamic Upsampler Module (DDUM), which processes edge-rich shallow features and semantic deep features in parallel before an attention-based fusion, avoiding feature contamination. An Information Compensation Module (ICM) further recovers fine details using multi-dilation convolutions. For challenging low-contrast Inner Vascular Complex (IVC) images, we introduce a multimodal fusion strategy, where a Feature Alignment Module (FAM) aligns probability maps from auxiliary modalities to enhance the IVC representation. Extensive experiments on five public datasets demonstrate that DDU-Net achieves state-of-the-art performance. Rigorous Wilcoxon signed-rank tests confirm these improvements are statistically significant, establishing DDU-Net as a reliable new baseline for quantitative clinical analysis. The code for DDU-Net is available on GitHub (https://github.com/steve706/DDUM_final).

The study of inter-beat intervals (IBI) and systolic blood pressure (SBP) fluctuations is of public health importance. Here we obtain insights about their underlying dynamics by means of an innovative study of the distribution of their rank-ordered registers, provided by fits to the Discrete Generalized Beta Distribution (DGBD), for healthy subjects and patients with end-stage renal disease (ESRD), under an active standing maneuver. SBP and IBI non-invasive time series were recorded during supine position followed by active standing for nine ESRD patients and eighteen age-matched healthy subjects. Once the data were rank ordered, the three parameter DGBD function was fitted through the Levenberg-Marquardt non-linear algorithm. Taking into consideration the statistical interpretations of the parameters, the quantitative exploration of their dependence with regard to the cases examined and changes in body position provided new insights: (i) Evidence for the presence of regulatory mechanisms that preserve the tail symmetry of the IBI distributions in healthy subjects, which are not evident in ESRD patients; (ii) The identification of a more pronounced weight of low-magnitude fluctuations at active standing in the SBP time series, manifested as a broader statistical dispersion of blood pressure values; (iii) A quantitative determination of a more undermined SBP regulation in ESRD. Overall, a better understanding of the statistical behavior of IBI and SBP time series is achieved by means of the DGBD function. Through the variation of its parameters, the DGBD approach has the potential to become a marker for assessing or even predicting the impairment of cardiovascular control mechanisms.