ObjectiveThe widespread adoption of portable fundus cameras for primary care and community screening is hindered by limitations in current autofocus(AF) technologies. Image-based methods relying on sharpness evaluation require iterative searches, resulting in slow convergence, while projection-based techniques are susceptible to optical artifacts and calibration errors. To address these challenges, this study introduces a novel AF system based on direct wavefront sensing, designed to deliver simultaneous high speed, high precision, and operational robustness within the compact form factor essential for portable ophthalmic devices. MethodsOur approach fundamentally reimagines the AF process by directly measuring the ocular wavefront aberration. We developed a custom portable fundus camera integrating a miniaturized Shack-Hartmann wavefront sensor (SHWS) into the optical path. An 850 nm laser diode projects a point source onto the retina via oblique illumination to minimize corneal reflections. Light scattered from this spot carries the eye’s refractive error through the imaging optics and is directed to the SHWS, positioned at a plane optically conjugate to the primary color CMOS imaging sensor. A microlens array within the SHWS samples the incident wavefront, generating a pattern of focal spots on a CCD. Real-time centroid analysis of these spots provides a map of local wavefront slopes. These measurements are processed through a singular value decomposition (SVD) algorithm to fit a Zernike polynomial basis set, enabling real-time reconstruction of the wavefront phase. The defocus component (S) is extracted from the second-order Zernike coefficients, providing a direct, quantitative measure of the refractive error in diopters. This value serves as a precise error signal in a closed-loop control system, which commands a voice-coil actuated focusing lens to its null position in a single, deterministic step, eliminating the need for iterative search algorithms. ResultsComprehensive evaluation demonstrated the system’s high performance. Testing on a calibrated model eye (OEMI-7) established a highly linear relationship between the computed defocus S and the focusing lens position across a ±20 Diopter (D) compensation range, achievable within a 5 mm mechanical travel. The system achieved a focusing precision of 0.08 D, corresponding to an 18-fold improvement over a conventional projection spot-size method tested under identical conditions. The total focus acquisition time, encompassing wavefront measurement, computation, and lens actuation, averaged under 0.5 s. Clinical validation with 25 human volunteers (50 eyes, refractive range -15 D to +10 D) confirmed practical efficacy. The wavefront-sensing AF succeeded in 92% of attempts with a mean time of 0.5 s, substantially outperforming a projection-based benchmark which achieved only a 32% success rate with an average time of 4.25 s. The system provided instantaneous directional guidance and maintained stability during minor ocular movements. Objective assessment of image quality, via amplitude contrast of retinal vasculature, showed consistent and significant enhancement following AF correction across the entire tested diopter range. ConclusionThis work successfully implements and validates a direct wavefront-sensing autofocus paradigm for portable fundus cameras. By directly quantifying and compensating for the optical defocus aberration, this method bypasses the fundamental limitations of image-processing and projection-based techniques, enabling rapid, precise, and deterministic diopter compensation. The developed system delivers an exceptional combination of a wide operational range (±20 D), high accuracy (0.08 D), fast convergence (0.5 s), and a compact physical footprint. This technology provides a practical and high-performance focusing solution capable of enhancing the reliability, throughput, and diagnostic utility of portable retinal imaging in large-scale screening applications. Future efforts will be directed towards system cost optimization and performance adaptation for diverse ocular conditions.

ObjectiveDiscriminating atypical hepatocellular carcinoma (HCC) from other malignancies in liver nodules classified as Liver Imaging Reporting and Data System category M (LR-M) remains a significant diagnostic challenge on conventional ultrasound examination. The LR-M category, originally intended to capture non-HCC malignancies, paradoxically contains up to 63% of atypical HCCs that deviate from classic enhancement patterns, leading to potential misdiagnosis and suboptimal treatment planning. While deep learning has shown promise in HCC diagnosis, most existing models rely exclusively on single-modality ultrasound, overlooking the diagnostic benefits of integrating complementary information from multiple imaging sources. To address this gap, we propose a novel attention-weighted tri-modal ultrasound network (TUS-Net) that integrates contrast-enhanced ultrasound (CEUS), B-mode ultrasound (BUS), and time-intensity curves (TICs) to improve diagnostic accuracy for these clinically challenging lesions. MethodsOur framework incorporates a three-dimensional convolutional neural network (C3D) backbone to extract spatiotemporal features from CEUS videos, capturing dynamic vascular patterns critical for lesion characterization. To effectively fuse complementary modalities, we introduce a dual-channel feature fusion module (DCFFM) that adaptively combines features from CEUS and BUS through channel-wise attention mechanisms, allowing the model to dynamically weigh the contribution of each modality based on diagnostic relevance. Additionally, we propose a temporal intensity feature fusion module (TIFFM) that leverages quantitative hemodynamic information from TICs to guide the model’s attention toward diagnostically critical temporal phases, such as arterial wash-in and portal venous washout. The model is further enhanced by automated lesion localization using YOLOX and class activation mapping for interpretability, ensuring that predictions align with clinically meaningful imaging features. ResultsEvaluated on a tri-modal ultrasound dataset comprising 161 patients with pathologically confirmed LR-M nodules (131 atypical HCC and 30 non-HCC malignancies), our model achieved an accuracy of 86.83%, a sensitivity of 92.50%, a specificity of 75.50%, and an AUC of 89.32% in screening atypical HCC. Compared to single-modality baselines, TUS-Net demonstrated superior specificity, a clinically critical metric given the higher risk associated with misclassifying non-HCC malignancies. Ablation studies confirmed the contribution of each module, with the full model outperforming both standard C3D and 3D ResNet backbones integrated with attention mechanisms. A reader study involving junior and senior radiologists further validated the clinical utility of AI assistance, showing consistent improvements in specificity and inter-reader consistency, particularly for less experienced clinicians. ConclusionThese results surpass existing benchmark models and demonstrate the potential of our approach to enhance diagnostic precision in clinically specific cases. By intelligently fusing multi-modal ultrasound data with attention-guided mechanisms, TUS-Net offers a reliable and interpretable tool that holds promise for improving the non-invasive diagnosis of atypical HCC in challenging LR-M liver nodules.

ObjectiveDiscriminating atypical hepatocellular carcinoma (HCC) from other malignancies in liver nodules classified as Liver Imaging Reporting and Data System category M (LR-M) remains a significant diagnostic challenge on conventional ultrasound examination. The LR-M category, originally intended to capture non-HCC malignancies, paradoxically contains up to 63% of atypical HCCs that deviate from classic enhancement patterns, leading to potential misdiagnosis and suboptimal treatment planning. While deep learning has shown promise in HCC diagnosis, most existing models rely exclusively on single-modality ultrasound, overlooking the diagnostic benefits of integrating complementary information from multiple imaging sources. To address this gap, we propose a novel attention-weighted tri-modal ultrasound network (TUS-Net) that integrates contrast-enhanced ultrasound (CEUS), B-mode ultrasound (BUS), and time-intensity curves (TICs) to improve diagnostic accuracy for these clinically challenging lesions. MethodsOur framework incorporates a three-dimensional convolutional neural network (C3D) backbone to extract spatiotemporal features from CEUS videos, capturing dynamic vascular patterns critical for lesion characterization. To effectively fuse complementary modalities, we introduce a dual-channel feature fusion module (DCFFM) that adaptively combines features from CEUS and BUS through channel-wise attention mechanisms, allowing the model to dynamically weigh the contribution of each modality based on diagnostic relevance. Additionally, we propose a temporal intensity feature fusion module (TIFFM) that leverages quantitative hemodynamic information from TICs to guide the model’s attention toward diagnostically critical temporal phases, such as arterial wash-in and portal venous washout. The model is further enhanced by automated lesion localization using YOLOX and class activation mapping for interpretability, ensuring that predictions align with clinically meaningful imaging features. ResultsEvaluated on a tri-modal ultrasound dataset comprising 161 patients with pathologically confirmed LR-M nodules (131 atypical HCC and 30 non-HCC malignancies), our model achieved an accuracy of 86.83%, a sensitivity of 92.50%, a specificity of 75.50%, and an AUC of 89.32% in screening atypical HCC. Compared to single-modality baselines, TUS-Net demonstrated superior specificity, a clinically critical metric given the higher risk associated with misclassifying non-HCC malignancies. Ablation studies confirmed the contribution of each module, with the full model outperforming both standard C3D and 3D ResNet backbones integrated with attention mechanisms. A reader study involving junior and senior radiologists further validated the clinical utility of AI assistance, showing consistent improvements in specificity and inter-reader consistency, particularly for less experienced clinicians. ConclusionThese results surpass existing benchmark models and demonstrate the potential of our approach to enhance diagnostic precision in clinically specific cases. By intelligently fusing multi-modal ultrasound data with attention-guided mechanisms, TUS-Net offers a reliable and interpretable tool that holds promise for improving the non-invasive diagnosis of atypical HCC in challenging LR-M liver nodules.

3CL protease (3CLpro) of SARS-CoV-2 is a pivotal enzyme required in coronavirus replication and transcription. Its highly conserved structure and the absence of homologous proteins in the host make it an ideal target for broad-spectrum anti-coronavirus drug development. In this work, we systematically investigated and compared the binding modes and dynamic properties of the four stereoisomers of a pyridyl-urea diyne ester (PyDU) molecule with two chiral centers within the 3CLpro active site. Through molecular docking, MD simulations, MM/GBSA binding free-energy calculations, and DCCM analysis, all four stereoisomers were stabilized primarily by hydrophobic packing. Among them, the (R,S) isomer exhibited the best overall performance, including docking score, binding free-energy components, and key residue interactions. The (R,S) and (S,R) isomers also enhanced the cooperative motions around the binding pocket, while the (R,S) isomer further modulated the flexibility of domain III, which may influence 3CLpro dimerization. Conversely, the (S,S) isomer exhibited the weakest affinity due to insufficient hydrophobic contact. By innovatively integrating chirality, binding energy and protein dynamical features, we revealed the dual role of chirality in modulating affinity and dynamic responses, which provides a theoretical basis for the chiral-guided design of coronavirus inhibitors.

Owing to its inherent advantages—such as being non-destructive, rapid, highly molecule-specific, and minimally interfered with by moisture—Raman spectroscopy has been widely adopted in the fields of skin barrier function assessment, monitoring the transdermal penetration of active cosmetic ingredients, and the identification and quality control of cosmetic products. Despite these strengths, the practical application of this technique faces a significant bottleneck: the strong fluorescence background generated by endogenous skin components and exogenous cosmetic additives. Endogenous skin substances, such as structural proteins (e.g., collagen and elastin), metabolic coenzymes (e.g., nicotinamide adenine dinucleotide), and pigments (e.g., melanin), together with exogenous cosmetic constituents like organic colorants, chemical sunscreens, and fragrances, often possess strong absorption and emission characteristics. When excited by lasers, these components produce a fluorescence background that can be 10⁶ to 10⁸ times stronger than the Raman scattering signals, effectively masking the inherently weak vibrational fingerprint information. In recent years, driven by the rapid development of optoelectronic hardware and artificial intelligence algorithms, fluorescence suppression strategies have evolved from isolated, single-method approaches into comprehensive, multi-level synergistic systems. These systems are categorized into three distinct tiers: sample preparation, signal acquisition, and data processing. At the sample preparation level, techniques such as photobleaching and surface-enhanced Raman spectroscopy (SERS) are employed to eliminate or bypass the generation of fluorescence at the source. At the signal acquisition level, instrumental improvements—including the use of long-wavelength near-infrared excitation (typically 785 nm or 1 064 nm), confocal spatial filtering, and shifted excitation Raman difference spectroscopy (SERDS)— are utilized to physically isolate Raman signals from the fluorescence background. Furthermore, at the data processing level, numerical baseline correction methods such as polynomial fitting, penalized least squares (e.g., airPLS, arPLS), wavelet transform, and derivative algorithms are increasingly integrated into the analytical pipeline to extract Raman spectral features from mixed signals without increasing hardware costs or acquisition time. This review provides a systematic categorization and critical evaluation of these fluorescence suppression methods, detailing their underlying principles, technical advantages, and inherent limitations in diverse experimental setups. By focusing on critical application scenarios—including skin barrier assessment, percutaneous absorption monitoring, the routine quality control of cosmetics, and the emerging field of portable on-site detection—this paper explores the current state of technique selection and optimization. Finally, the article discusses future development trends, emphasizing the necessity of constructing adaptive, tiered suppression strategies, developing intelligent and automated data processing algorithms, and promoting the integration of portable, multi-modal diagnostic devices. The objective of this review is to provide a comprehensive technical reference to facilitate the transition of Raman spectroscopy from a specialized laboratory tool into a routine, robust analytical platform for advancements in skin science and cosmetic research.

Circadian rhythm is an endogenous biological clock mechanism that enables organisms to adapt to the earth’s alternation of day and night. It plays a fundamental role in regulating physiological functions and behavioral patterns, such as sleep, feeding, hormone levels and body temperature. By aligning these processes with environmental changes, circadian rhythm plays a pivotal role in maintaining homeostasis and promoting optimal health. However, modern lifestyles, characterized by irregular work schedules and pervasive exposure to artificial light, have disrupted these rhythms for many individuals. Such disruptions have been linked to a variety of health problems, including sleep disorders, metabolic syndromes, cardiovascular diseases, and immune dysfunction, underscoring the critical role of circadian rhythm in human health. Among the numerous systems influenced by circadian rhythm, the skin—a multifunctional organ and the largest by surface area—is particularly noteworthy. As the body’s first line of defense against environmental insults such as UV radiation, pollutants, and pathogens, the skin is highly affected by changes in circadian rhythm. Circadian rhythm regulates multiple skin-related processes, including cyclic changes in cell proliferation, differentiation, and apoptosis, as well as DNA repair mechanisms and antioxidant defenses. For instance, studies have shown that keratinocyte proliferation peaks during the night, coinciding with reduced environmental stress, while DNA repair mechanisms are most active during the day to counteract UV-induced damage. This temporal coordination highlights the critical role of circadian rhythms in preserving skin integrity and function. Beyond maintaining homeostasis, circadian rhythm is also pivotal in the skin’s repair and regeneration processes following injury. Skin regeneration is a complex, multi-stage process involving hemostasis, inflammation, proliferation, and remodeling, all of which are influenced by circadian regulation. Key cellular activities, such as fibroblast migration, keratinocyte activation, and extracellular matrix remodeling, are modulated by the circadian clock, ensuring that repair processes occur with optimal efficiency. Additionally, circadian rhythm regulates the secretion of cytokines and growth factors, which are critical for coordinating cellular communication and orchestrating tissue regeneration. Disruptions to these rhythms can impair the repair process, leading to delayed wound healing, increased scarring, or chronic inflammatory conditions. The aim of this review is to synthesize recent information on the interactions between circadian rhythms and skin physiology, with a particular focus on skin tissue repair and regeneration. Molecular mechanisms of circadian regulation in skin cells, including the role of core clock genes such as Clock, Bmal1, Per and Cry. These genes control the expression of downstream effectors involved in cell cycle regulation, DNA repair, oxidative stress response and inflammatory pathways. By understanding how these mechanisms operate in healthy and diseased states, we can discover new insights into the temporal dynamics of skin regeneration. In addition, by exploring the therapeutic potential of circadian biology in enhancing skin repair and regeneration, strategies such as topical medications that can be applied in a time-limited manner, phototherapy that is synchronized with circadian rhythms, and pharmacological modulation of clock genes are expected to optimize clinical outcomes. Interventions based on the skin’s natural rhythms can provide a personalized and efficient approach to promote skin regeneration and recovery. This review not only introduces the important role of circadian rhythms in skin biology, but also provides a new idea for future innovative therapies and regenerative medicine based on circadian rhythms.

ObjectivePrimary liver cancer, predominantly hepatocellular carcinoma (HCC), is a significant global health issue, ranking as the sixth most diagnosed cancer and the third leading cause of cancer-related mortality. Accurate and early diagnosis of HCC is crucial for effective treatment, as HCC and non-HCC malignancies like intrahepatic cholangiocarcinoma (ICC) exhibit different prognoses and treatment responses. Traditional diagnostic methods, including liver biopsy and contrast-enhanced ultrasound (CEUS), face limitations in applicability and objectivity. The primary objective of this study was to develop an advanced, light-weighted classification network capable of distinguishing HCC from other non-HCC malignancies by leveraging the automatic analysis of brightness changes in CEUS images. The ultimate goal was to create a user-friendly and cost-efficient computer-aided diagnostic tool that could assist radiologists in making more accurate and efficient clinical decisions. MethodsThis retrospective study encompassed a total of 161 patients, comprising 131 diagnosed with HCC and 30 with non-HCC malignancies. To achieve accurate tumor detection, the YOLOX network was employed to identify the region of interest (ROI) on both B-mode ultrasound and CEUS images. A custom-developed algorithm was then utilized to extract brightness change curves from the tumor and adjacent liver parenchyma regions within the CEUS images. These curves provided critical data for the subsequent analysis and classification process. To analyze the extracted brightness change curves and classify the malignancies, we developed and compared several models. These included one-dimensional convolutional neural networks (1D-ResNet, 1D-ConvNeXt, and 1D-CNN), as well as traditional machine-learning methods such as support vector machine (SVM), ensemble learning (EL), k-nearest neighbor (KNN), and decision tree (DT). The diagnostic performance of each method in distinguishing HCC from non-HCC malignancies was rigorously evaluated using four key metrics: area under the receiver operating characteristic (AUC), accuracy (ACC), sensitivity (SE), and specificity (SP). ResultsThe evaluation of the machine-learning methods revealed AUC values of 0.70 for SVM, 0.56 for ensemble learning, 0.63 for KNN, and 0.72 for the decision tree. These results indicated moderate to fair performance in classifying the malignancies based on the brightness change curves. In contrast, the deep learning models demonstrated significantly higher AUCs, with 1D-ResNet achieving an AUC of 0.72, 1D-ConvNeXt reaching 0.82, and 1D-CNN obtaining the highest AUC of 0.84. Moreover, under the five-fold cross-validation scheme, the 1D-CNN model outperformed other models in both accuracy and specificity. Specifically, it achieved accuracy improvements of 3.8% to 10.0% and specificity enhancements of 6.6% to 43.3% over competing approaches. The superior performance of the 1D-CNN model highlighted its potential as a powerful tool for accurate classification. ConclusionThe 1D-CNN model proved to be the most effective in differentiating HCC from non-HCC malignancies, surpassing both traditional machine-learning methods and other deep learning models. This study successfully developed a user-friendly and cost-efficient computer-aided diagnostic solution that would significantly enhances radiologists’ diagnostic capabilities. By improving the accuracy and efficiency of clinical decision-making, this tool has the potential to positively impact patient care and outcomes. Future work may focus on further refining the model and exploring its integration with multimodal ultrasound data to maximize its accuracy and applicability.

With the continuous development of society, a large number of new chemicals are continuously emerging, which presents a challenge to current risk assessment and safety management of chemicals. Traditional toxicology research methods have certain limitations in quickly, efficiently, and accurately assessing the toxicity of many chemicals, and cannot meet the actual needs. In response to this challenge, computational toxicology that use mathematical and computer models to achieve the prediction of chemical toxicity has emerged. In the meantime, as researchers increasingly pay attention to understanding the interaction mechanisms between exogenous chemical substances and the body from the system level, and multiomics technologies develop rapidly such as genomics, transcriptomics, proteomics, and metabolomics, huge amounts of data have been generated, providing rich information resources for studying the interactions between chemical substances and biological molecules. System toxicology and network toxicology have also developed accordingly. Of these, network toxicology can integrate these multiomics data to construct biomolecular networks, and then quickly predict the key toxicological targets and pathways of chemicals at the molecular level. This paper outlined the concept and development of network toxicology, summarized the main methods and supporting tools of network toxicology research, expounded the application status of network toxicology in studying potential toxicity of exogenous chemicals such as agricultural chemicals, environmental pollutants, industrial chemicals, and foodborne chemicals, and analyzed the development prospects and limitations of network toxicology research. This paper aimed to provide a reference for the application of network toxicology in other fields.

Objective:To develop an artificial intelligence (AI)-based platelet review strategy to identify abnormal platelet histograms with no significant difference between initial impedance platelet count (PLT-I) and PLT-F results.Methods:This study included 5 119 routine blood analysis in Nanfang Hospital of Southern Medical University and its Ganzhou branch from July 2023 and March 2024. Specimens exhibiting abnormal platelet histograms and an initial platelet count >40×10?/L underwent review using the fluorescent platelet count (PLT-F) channel. Consistency of the results was defined as a difference between impedance platelet count (PLT-I) and PLT-F less than ±20% of the PLT-F results. A deep learning model was developed using platelet and red blood cell histogram data from a training set of 3 807 specimens. The model′s diagnostic performance was evaluated on an independent external validation set ( n=805) using receiver operating characteristic (ROC) curve analysis. Changes in the number of reviewed samples and sample turnaround time were analyzed to assess its clinical utility. Results:The deep learning model based on platelet and red blood cell histograms achieved an area under the ROC curve (AUC) of 0.854 in the training set. At a cutoff value of 0.1, the sensitivity was 0.954 and specificity was 0.358. The model could reduce review by 16.80% (190/1 131). In the validation set, the AUC was 0.805, with a sensitivity of 0.955 and specificity of 0.307, corresponding to a reduction of 17.41% (47/270) in reviewed specimens.Conclusion:The platelet review prediction model developed based on deep learning technology can efficiently identify samples with consistent results before and after review, reducing unnecessary reviews and shortening specimen testing time, thereby improving the efficiency of platelet test.