Abstract:An overview is provided of the research progress in the application of hyperspectral detection technology for non-destructive testing of key parameters in tobacco leaf quality. Methods and equipment for the rapid detection of chemical components such as total sugar， reducing sugar， total nitrogen， nicotine， starch， chloride， and potassium in tobacco leaves using this technology are explored. The impact of different tobacco sample forms on spectral data is pointed out. The advantages and challenges of hyperspectral technology in applications such as field management， harvest optimization， and online grading in tobacco production are analyzed. The promising prospects of combining hyperspectral technology with artificial intelligence to build predictive models for tobacco leaf chemical composition are proposed. This combination provides scientific evidence and references for improving detection efficiency and quality in the tobacco industry.

Abstract:Enamel demineralization often occurs in the early stage of dental caries. Studying the microscopic mechanism of enamel demineralization is essential to prevent and treat dental caries. Terahertz （THz） technology， especially continuous wave （CW） THz near-field scanning microscopy （THz-SNOM） with its nanoscale resolution， can be promising in biomedical imaging. In addition， compared with traditional THz time-domain spectroscopy （TDS）， portable solid-state source as the emission has higher power and SNR， lower cost， and can obtain more precise imaging. In this study， we employ CW THz-SNOM to further break the resolution limitations of conventional THz imaging techniques and successfully achieve the near-field imaging of demineralized enamel at the nanoscale. We keenly observe that the near-field signal of the enamel significantly lowers as demineralization deepens， mainly due to the decrease in permittivity. This new approach offers valuable insights into the microscopic processes of enamel demineralization， laying the foundation for further research and treatment.

Abstract:Optical Coherence Tomography （OCT） provides high-resolution images of skin tissue structure and pathological features. Automated image analysis methods （such as segmentation and classification） are important for assisting skin disease diagnosis and treatment evaluation. These methods provide quantitative support for medical decisions. Compared with traditional methods and early machine learning （ML） techniques， deep learning （DL） improved analysis efficiency and reproducibility. It also reduced manual processing time significantly. This paper systematically reviewed the application progress of DL in skin OCT image analysis. It focused on technical approaches for image denoising， skin layer segmentation， and skin cancer diagnosis. The study identified key challenges including model generalization and data heterogeneity. The findings provide theoretical references and technical guidance for future research directions.

Abstract:High spatiotemporal resolution multi-region brain synchronization imaging is a critical requirement in neural circuit research. However, traditional multiphoton microscopy is limited by its single field-of-view (FOV) imaging mode, making it difficult to achieve large-scale observation of neural activity across multiple brain regions. The multi-FOV multi-photon imaging technology employs a FOV segmentation strategy in both the front and rear optical paths of the objective lens and combines multi-dimensional signal analysis methods (such as wavelength encoding, spatial demultiplexing, and time gating) to effectively overcome the spatiotemporal resolution limitations of traditional techniques. This technology enables millisecond-level temporal resolution and micron-level spatial resolution for synchronous imaging across brain regions, providing a novel research paradigm for revealing cortical functional coupling, cortical-subcortical neural circuit coordination mechanisms, and whole-brain neural signal propagation dynamics. In the future, through in-depth integration with techniques such as endoscopic imaging, adaptive optical aberration correction, optical stimulation and deep learning-based image analysis, multi-FOV multi-photon imaging will further advance the precise decoding of neural circuit functional architecture and demonstrate significant value in clinical translation fields such as neurodegenerative disease diagnosis and brain-machine interface development.

Abstract:The concentration of exhaled CO as a biomarker for certain diseases has attracted significant attention. However， existing CO concentration detectors suffer from low sensitivity and slow response times. To address this， we developed a high-sensitivity， rapid-response exhaled CO measurement system based on absorption spectroscopy， utilizing a quantum cascade laser with a central wavelength of 4.59 μm and a 3.8 m multi-pass cell. The CO concentration was analyzed using both direct absorption spectroscopy （DAS） and wavelength modulation spectroscopy （WMS）. The DAS method demonstrated a linearity of 0.998 with a detection limit of 3.68 × 10??. For WMS， the linearity remained 0.998 at CO concentrations below 6.00 × 10??， achieving a detection limit of 3.00 × 10??. Through Allan variance analysis， optimal integration times of 170 s for DAS and 250 s for WMS were determined， corresponding to improved detection limits of 2.00 × 10?? and 3.00 × 10?1?， respectively. Finally， exhaled CO concentrations from 14 volunteers were measured， demonstrating the system"s capability to distinguish between smokers and non-smokers. This provides a scientifically validated tool for assessing smoking status in clinical smoking cessation programs.

Abstract:Fibroblasts support a broad range of essential organ functions via microarchitectural， biomechanical， and biochemical cues. Despite great advances in fluorescence， photoacoustic conversion， and Raman scattering over the past decades， their invasiveness and limited spatial resolution hinder the characterization of fibroblasts in a single cell. Here， taking mouse embryonic fibroblasts （MEFs） as an example， we propose a novel noninvasive approach to investigate the compositional distribution of MEFs at the single-cell scale via terahertz （THz） nanoscopy. Compared to the topological morphology， THz nano-imaging enables the component-based visualization of MEFs， such as the membrane， cytoplasm， nucleus， and extracellular vesicles （EVs）. Notably， we demonstrate the real-space observation of the influence of rapamycin treatment on the increase of EVs in MEFs. Moreover， the line-cut and area-statistical analysis establishes the relationship between the topological morphology and the THz near-field amplitudes for different cellular components of MEFs. This work provides a new pathway to characterize the effects of pharmaceutical treatments， with potential applications in disease diagnosis and drug development.

Abstract:Near-infrared spectroscopy is a type of molecular vibration spectroscopy. Temperature variations cause changes in molecular vibrations such as O-H and inter molecular forces such as hydrogen bonding， which lead to absorption spectral intensity and peaks changes， affecting the prediction accuracy of minor components such as blood glucose. To address the impact of temperature perturbation on spectral detection and modeling analysis， a method of temperature perturbation discrimination based on aquaphotomics and two-trace two-dimensional correlation spectroscopy （2T2D-COS） was proposed. The 2T2D-COS analysis was applied to diffuse reflectance spectra of simulated solutions under temperature perturbation and varying glucose concentrations. Spectral features induced by changes in temperature and glucose concentration were successfully extracted， revealing distinct water spectral patterns under different perturbations. Quantitative analysis shows that temperature changes of 0.1 °C is equivalent to glucose concentration changes of 45 mg/dL in terms of intensity. A temperature perturbation outliers discrimination model was further established based on raw spectra， water spectral features， and 2T2D-COS asynchronous spectra. The accuracy rates of the model based on 2T2D-COS asynchronous spectra are 95.83%. After removing outliers， the root mean square error of glucose concentration prediction is reduced by 51.89%. This work provides a foundation for improving the accuracy of in vivo blood glucose detection using near-infrared spectroscopy.

Abstract:Traumatic brain injury is one of the most serious diseases that endanger human health. Sensitive and rapid detection method is a kind of powerful guarantee for the accurate and effective treatment of traumatic brain injury. In recent years， terahertz （THz） wave and Raman spectroscopy have broad application prospects in biomedical diagnosis and other fields due to their complementarity in technology. In this article， the researches of terahertz wave and Raman spectroscopy technology in traumatic brain injury detection were summarized in response to the needs and difficulties of traumatic brain injury diagnosis. Firstly， the development status of THz imaging and THz spectroscopy technology was introduced， and the applications of the two technologies in traumatic brain injury detection were also introduced， respectively. In addition， the principle and classification of Raman spectroscopy were summarized， and the research of Raman spectroscopy in the detection of traumatic brain injury tissues， body fluids， and biomarkers were discussed. Finally， the development trend of THz wave and Raman spectroscopy in the detection of traumatic brain injury was analyzed， which provides a new research idea for the application of THz wave and Raman spectroscopy in the rapid and accurate diagnosis of traumatic brain injury.

Abstract:Reflective tomography lidar (RTL) reconstructs target contours by acquiring laser echo projection data, but incomplete angular detection in practice often leads to insufficient projection data. To address this issue, the authors proposes a target contour reconstruction method that combines the structural sparsity of projection data with a super-resolution convolutional neural network (SRCNN), based on the principles and technical implementation of RTL. This approach effectively resolves the failure of traditional algorithms when projection data suffers from severe angular deficiency. Different from conventional RTL imaging methods that directly incorporate sparse reconstruction models, the authors first recovers full-angle projection data by integrating sparse constraints with SRCNN based on geometry prior of the projection data, followed by standard RTL imaging algorithms to achieve complete targets contour reconstruction. To validate the effectiveness of the proposed method, the authors designed laser echo projection simulations based on the facet model and conducted field experiments. The results demonstrate that the authors achieves high-quality target contour reconstruction under varying levels of projection data missing conditions.

Abstract:Central precocious puberty （CPP） is mainly caused by the premature activation of the hypothalamic-pituitary-gonadal axis， which leads to abnormal hormone levels and triggers structural and functional changes in the brain， making the neurovascular coupling mechanisms of children with CPP different from those of normal children in the task state. Addressing current limitations of clinical diagnosis， such as false negatives， interference from obesity， and physiological discomfort， this study utilized functional near-infrared spectroscopy （fNIRS） to analyze task-related brain activation characteristics in 167 children from Tianjin Hospital， including 85 normal children and 82 children with CPP. An auxiliary diagnostic model for CPP was established based on these analyses. It was found that the prefrontal activation areas during mental arithmetic （MA） were more in the normal group than in the CPP group， and the activation areas were more in females than in males. By selecting mean， variance， kurtosis， and skewness from the two channels with the highest frequency of correlation and the highest magnitude of negative correlation as input features， the constructed classification model achieved an accuracy rate of 79.1%. This study provides a new and important reference for the rapid screening and pathogenesis study of CPP.