Biomedical Physics & Engineering Express最新文献

Artificial intelligence (AI) systems are increasingly used in medical imaging for disease diagnosis, yet their vulnerability to adversarial attacks poses significant risks for clinical deployment. In this work, we systematically evaluate the susceptibility of VolBrain, a widely used third-party neuroimaging diagnostic platform, to universal black-box adversarial attacks. We generate adversarial perturbations using a surrogate convolutional neural network trained on a different dataset and with a different architecture, representing a worst-case scenario for the attacker where they have no access to the internals of the system. For this, we employ both the Fast Gradient Sign Method (FGSM) and DeepFool attacks. Our results show that these perturbations can reliably degrade the diagnostic performance of VolBrain, with DeepFool-based attacks being particularly effective for comparable perturbation sizes. We further demonstrate that a simple Mean Attack approach is also effective in degrading VolBrain performance, showing that this system is vulnerable to universal attacks, that is, perturbations agnostic to the input. These findings highlight the substantial risk posed by universal black-box adversarial attacks, even when attackers lack access to the target model or its training data. Our study underscores the urgent need for robust defense mechanisms and motivates further research into the adversarial robustness of medical AI systems.

Objective. This study explores a more reliable method for measuring nociceptive pain induced by laser stimuli from electroencephalography (EEG) signals, addressing the limitations of fixed pain scales by incorporating inter-individual variability in subjective pain tolerance.Approach. For this purpose, a public database was used that includes recordings from 51 subjects who received controlled laser stimuli at three different intensities on the back of the hand to evoke pain, while EEG activity was simultaneously recorded. Signal processing techniques were then applied to extract power in six frequency bands (e.g., alpha, beta, gamma). The extracted features were fed into machine learning algorithms to predict pain levels. This prediction was performed by comparing two data labeling strategies (reaction time versus laser intensity) and two different EEG channel configurations (62 channels versus 20 somatosensory channels).Main results. The power of EEG frequency bands, combined with machine learning, distinguished pre-stimulus from in-stimulus conditions with an average accuracy of 86%. Classification across pain levels was more challenging, reaching a maximum of 63% in the binary discrimination between high and low pain. The 62-channel configuration and the 20-channel somatosensory setup showed similar performance, although in some cases the 62-channel setup yielded better results. Incorporating temporal information from reaction times further improved performance, with time-based labels significantly outperforming intensity-based labels.Significance. Our results indicate that the best labeling system for predicting nociceptive pain levels is that one based on reaction time (p-value < 0.001; two-sided Student's t-test), thus suggesting that pain perception is subjective and that classifying pain solely based on stimulus intensity may not be reliable.

Electrocardiogram (ECG) and phonocardiogram (PCG) have emerged as crucial non-invasive and portable diagnostic modalities for early cardiovascular disease (CVD) screening. Despite the individual merits of these signal modalities in CVD detection, significant challenges persist, including insufficient inter-modal interaction and suboptimal weight allocation. To address these critical limitations, we proposed a novel Time-Frequency Cross-Modal Attention Fusion Network (TF-CrossNet) designed for precise early CVD diagnosis. The proposed network employs a dual-path multiscale residual structure to extract key time-frequency domain features from PCG and ECG signals, comprehensively capturing multiscale information. Leveraging the intrinsic electro-mechanical coupling relationship of the heart, a bidirectional mutual enhancement attention module is introduced to capture interactive morphological information between PCG and ECG signals, enabling feature-level signal complementation and enhancement. Furthermore, an adaptive fusion strategy based on Bayesian decision theory is developed, establishing a mapping relationship between confidence levels and loss functions to dynamically optimize modal weight allocation. Validated on the 2016 PhysioNet/CinC dataset, the model achieved exceptional performance metrics: 93.13% accuracy, 97.7% specificity, and 98% area under the curve (AUC). Furthermore, comprehensive noise robustness experiments demonstrate that TF-CrossNet maintains superior performance under various noise conditions, achieving an average robustness index of 94.20% compared to existing methods, validating its practical applicability in clinical environments. The superior effectiveness of the proposed approach in CVD classification, providing a novel technological pathway for non-invasive and precision CVD diagnosis.

Accurate segmentation of retinal vessels is critical for the diagnosis of ophthalmic diseases. However, this task is made challenging by two issues: vast-scale variations from major arteries to fine capillaries often lead to a fractured vessel topology, and low-contrast boundaries corrupted by noise frequently result in segmentation ambiguity. To address these challenges, we propose an adaptive spatial channel fusion high-resolution network (ASCF-HRNet). The proposed architecture has two synergistic innovations: first, to preserve the topological integrity of the vascular network against vast-scale variations, we propose a spatial semantic enhancement (SSE) block that replaces standard convolutions with parallel multi-scale kernels and spatial attention; and second, to resolve segmentation ambiguity at low-contrast boundaries, we design a channel feature enhancement (CFE) block. Strategically integrated prior to each upsampling operation, the features were purified by performing a semantics-aware refinement that prevented the propagation of background noise and redundant information. Extensive experiments on the DRIVE, CHASE_DB1, and STARE datasets demonstrate that ASCF-HRNet achieves leading AUC scores of 0.9880, 0.9899, and 0.9828, and highly competitive F1-scores of 0.8263, 0.8119, and 0.7781. The results demonstrate that our proposed ASCF-HRNet achieves a superior segmentation performance, particularly in preserving vascular topology and ensuring boundary fidelity.

Background. The appearance time of blood components in the brain provides complementary information about cerebral microvascular dynamics. Plasma and red blood cells (RBCs) behave differently in the microcirculation: while plasma can pass through peripheral layers of capillaries, RBCs carry oxygen and are affected by phenomena such as the Fåhræus-Lindqvist effect and plasma skimming. Our study aims at visualizing these differencesin vivousing H₂¹⁵O PET to assess plasma appearance time (AT_Plasma) and¹⁵O₂PET to assess RBC appearance time (AT_RBC), and demonstrated that the relative delay of RBCs compared with plasma correlates with the oxygen extraction fraction (OEF).Methods. We retrospectively analyzed PET images obtained with¹⁵O₂and H₂¹⁵O administration in 40 patients, comprising a total of 63 scan data. Appearance time images were generated by fitting tissue curves both for the¹⁵O₂(AT_RBC) and H₂¹⁵O phases (AT_Plasma). AT_RBCand AT_Plasmavalues were extracted from regions of interest (ROIs) and compared. Additionally, differences between AT_RBCand AT_Plasma(ΔAT) were analyzed in relation to OEF.Results. AT_RBCand AT_Plasmaimages exhibited similar spatial distributions. A strong correlation was observed between them as AT_RBC= 0.80·AT_Plasma+ 1.8,r= 0.86), with a slope significantly less than unity, suggesting that RBCs flow faster than plasma. The difference between the two (ΔAT) showed a moderate correlation with OEF (r= 0.44), suggesting that higher OEF values are associated with slower RBC movement relative to plasma. This finding suggests that under certain ischemic conditions, RBC flow is more severely impaired than plasma flow.Conclusion. This study demonstrates that AT_RBCand AT_Plasmaare closely related measures of cerebral blood appearance time. The observed association between their difference and OEF suggests a potential link to ischemic pathology.

Remote photoplethysmography (rPPG) offers a non-contact method for monitoring physiological signals using camera-based systems. The goal of this research is to estimate heart rate and spatial distributions of vascular perfusion using spatio-temporal rPPG (ST-rPPG) and to evaluate the impact of polarization, spectral filtering, and motion compensation on perfusion map quality and heart rate estimation. Two acquisition setups were used: an RGB camera with and without cross-polarization, and a monochrome camera combined with spectral filters. A motion compensation strategy was implemented that combined optical flow-based stable segment selection and temporal video stabilization to reduce motion artifacts. Four rPPG algorithms (GREEN, CHROM, POS, and G-R) were evaluated using three performance metrics: Absolute Error (AE), Signal Quality Index (SQI), and Signal-to-Noise Ratio (SNR) under cross polarized and non polarized lighting in 20 subjects to assess their suitability for perfusion mapping. GREEN and G-R method stood out giving the best results. In the second setup, nine spectral filters were tested across three anatomical regions using the GREEN method, to investigate the influence of wavelength selection on spatial perfusion signal quality. Green, orange, and blue wavelengths produced the best results in terms of AE, SQI and SNR, particularly in the palm region. Visualizations like the spatial perfusion maps, confirmed the superiority of motion-compensated, polarized, and spectrally optimized conditions for enhancing non-contact vascular perfusion assessment. Prior rPPG studies focused primarily on facial datasets or single optical factors, while this work provides the systematic evaluation of polarization, spectral filtering, and motion compensation in a unified hand-based framework, extending established rPPG methods toward high-resolution perfusion mapping.

Cross-section data unavailability for non-water materials in track structure simulation software necessitates nanodosimetric quantity transformation from water to other materials. Cluster dose calculation transformation initially employed mass-density-based scaling - an approach resulting in a physically unrealistic material-independence of the cluster dose equation. This study introduces an alternative scaling method based on material-specific ionization cross-sections. The mean free path ratio of the materials for both the primary particles of the track structure simulation and for the secondary electrons served as the scaling factor. The approach was demonstrated through a cluster dose calculation for a carbon ion beam in a realistic head geometry and compared to the previous scaling method. The proposed cross-section-based scaling method resulted in a physically expected increase in cluster dose values for denser materials, which was not visible in the original scaling approach. The introduced scaling approach can be used to determine cluster dose distributions in heterogeneous geometries, a fundamental requirement for its integration into radiotherapy treatment planning frameworks.

Retinal diseases are the leading causes of visual impairment, and early diagnosis is essential for treatment. Optical coherence tomography (OCT), a non-invasive imaging technique, provides high-resolution images for retinal disease classification; however, its complexity and the limitations of manual diagnosis require efficient automated classification methods. To enableearly clinical diagnosis, this paper proposes a Convolutional Block Attention Module (CBAM)- Automatic Mixing Precision (AMP) network (CANet) for automated classification of retinal images. The model categorizes diabetic macular oedema (DME), choroidal neovascularisation (CNV), drusen. and normal cases, which is built upon the ResNet-34 architecture. CBAM is introduced into the residual block to propose the CBAM-Block residual block embedded in the ResNet-34 network model, which combines the channel and spatial attention mechanism to enhance the lesion feature extraction capability. AMP is used to accelerate the training and combine with transfer learning to enhance the model generalization. Meanwhile, median filtering, normalization, dynamic thresholding to remove white edges and data enhancement are used to optimize data quality and alleviate the problem of category imbalance. Classification experiments were performed on the OCT-2017 dataset for the four categories and ablation experiments were performed to demonstrate their effectiveness. The total classification accuracy of the model reaches 0.9890, the AUC value of all categories is 1, where the recall of CNV reaches 1. CBAM, AMP, and transfer learning improve the classification accuracy by 0.9%, 1.6%, and 9.4%, respectively, and the ablation experiments likewise prove that the model remains highly robust to noisy data. The experimental results show that the CANet model significantly improves the OCT image classification performance through multi-module integration, which provides an efficient and reliable technical solution for the automated diagnosis of retinal diseases.

Background: PET is a highly sensitive imaging modality for visualizing metabolic processes.Objective: This study evaluates PET scanner designs using monolithic-like BGO detector crystals, aimed at enhancing sensitivity while having minimal impact on spatial resolution.Methods: Two PET scanners with 16 detector heads were simulated using the GATE: (1) a total-body (T-B) scanner with a 105cm axial field of view (AFOV), and (2) a whole-body (W-B) scanner with a 35cm AFOV. Both designs employed 1 × 1 × 1.6cm³BGO monolithic-like crystals. The performance of both scanners was assessed according to NEMA NU-2 2018 standards, including sensitivity, scatter fraction, NECR, and spatial resolution, and was compared with existing scanners. Additionally, point source sensitivity at the center of the scanner was compared with an analytical model to validate the simulation results.Results: A good agreement was observed between simulated and analytical point source sensitivities, with a maximum deviation of 4%. The T-B and W-B scanners achieved sensitivities of 39.73 and 17.87kcpsMBqat the center of the FOV. Scatter fractions were 35.5% and 29.1% for the T-B and W-B scanners, respectively. The NECR peak was 3498.2 kcps at ∼21kBqmlfor the T-B scanner, and 286.8 kcps at ∼14kBqmlfor the W-B scanner. Both scanners demonstrated average spatial resolutions of 2.66 mm (T-B) and 2.39mm (W-B) at the center of the scanner. At the center of the FOV, the T-B scanner showed 24% and 41.8% higher sensitivity compared to the Biograph-Vision Quadra and Walk-through PET scanners, respectively. Additionally, the W-B scanner showed 8.3% higher sensitivity at the center compared to the Biograph-Vision. The T-B and W-B scanners achieved 23% and 37.5% better spatial resolution at the scanner center compared to Biograph-Vision Quadra and Biograph-Vision, respectively.Conclusions: The proposed PET scanners with monolithic-like BGO crystals showed promising sensitivity and resolution, indicating improved PET imaging potential.

Objective.Nowadays the linear-quadratic model (LQ) is the most used model to estimate the biological effective dose (BED) and the equivalent dose in 2 Gy fractions (EQD₂) for different fractionation regimens. Nevertheless, it is debated of applicability to use LQ model for hypofractionation. The objective of this study is to evaluate the LQ model in comparison with other radiobiological models concerning the adequacy of biological equivalent dose in 2 Gy fractions assessment across various hypofractionation regimens.Methods.The study was conducted for two cases: the prostate gland in the pelvic region and squamous cell carcinoma (SCC) in the head and neck region. Five radiobiological models including the LQ model, modified linear-quadratic (MLQ), linear-quadratic-linear (LQL), universal survival curve (USC), and Pade linear-quadratic (PLQ) models were compared for tumor control probability (TCP) andEQD₂predictions. Published clinical outcomes (including local control, disease-free survival, and overall survival rates) were analyzed to identify clinically equivalent fractionation regimens. The radiobiological models were then evaluated by comparing calculatedEQD₂andTCPvalues with clinical data for these equivalent regimens.Results:Modified radiobiological models showed that the LQ model overestimates the dose in hypofractionation. The dose limit at which the LQ model is applicable depends on the localization and type of tumor: for the prostate gland the value was 4.3 Gy, for the head and neck region 8.5 Gy.Conclusions:The applicability of the LQ model in hypofractionation depends on the tumorα/βvalue: the LQ model more sensitive to locations with lowα/βvalues and, conversely, less sensitive to locations with highα/βvalues. Among the alternatives, the MLQ model is recognized as the most practical alternative, combining a small number of parameters with resistance to variations. While modified models show efficacy, further clinical validation is needed to balance tumor control with normal tissue toxicity risks.