A sepsis screening system and method based on hyperspectral imaging

Abstract

The application discloses a sepsis screening system and method based on hyperspectral imaging, and relates to the technical field of hyperspectral image processing.The sepsis screening system comprises a light source module, a double-channel multi-mode optical fiber transmission module, a data acquisition module and a hyperspectral push-broom imaging module.The light source module is used for emitting illumination light to the double-channel multi-mode optical fiber transmission module through a light source.The double-channel multi-mode optical fiber transmission module is used for transmitting the illumination light to the data acquisition module, and inputting the high-spectral signal obtained by converting the collected light signal into the hyperspectral push-broom imaging module.The data acquisition module is used for collecting the light signal of the tissue surface in real time.The hyperspectral push-broom imaging module is used for converting the high-spectral signal of the tissue surface into original cube data and transmitting the original cube data to the data processing module.The data processing module is used for processing the original cube data to obtain a sepsis identification result.The sepsis screening system overcomes the defects of the prior art, such as insufficient sensitivity, poor real-time performance, low flexibility of equipment and large amount of algorithm calculation, and realizes early and accurate screening of sepsis in real-time medical scenes such as ICU bedside.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral image processing technology, and in particular to a sepsis screening system and method based on hyperspectral imaging. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Sepsis is one of the most dangerous syndromes in the field of acute and critical care medicine, characterized by rapid progression and high mortality, placing extremely high demands on early, accurate, and continuous monitoring. However, the mainstream screening technologies currently relied upon by ICUs and emergency departments still rely primarily on traditional laboratory and imaging methods, forming an offline system with a "sampling-transportation-testing-interpretation" workflow. This system uses blood culture as the gold standard for pathogen identification, inflammatory markers such as white blood cell count, C-reactive protein, and procalcitonin as dynamic monitoring indicators, the SOFA score as a tool for quantifying organ dysfunction, and supplements anatomical assessment with CT imaging when necessary. While the aforementioned methods have long been recommended by guidelines, they generally suffer from inherent limitations such as low sensitivity and long reporting times: the positive rate of blood cultures is significantly affected by prior antibiotic use, sampling volume, and pathogen type, and the reporting time usually exceeds 24 hours; fluctuations in inflammatory factors lag behind pathophysiological changes, often showing a significant increase only when tissue hypoperfusion and cell damage are irreversible; multiple biochemical tests dependent on the SOFA score are also scattered across different time points, lacking real-time consistency; although CT scans can detect foci of infection and complications, they have high transport risks, many contraindications to contrast agents, and cannot continuously track microcirculatory changes. This resulting lack of a time window means that clinicians can only intervene at the "already occurred" stage rather than the "will occur" stage, missing the critical opportunity to block the sepsis cascade.

[0004] In the field of optical detection, to overcome the lag and discreteness of traditional methods, researchers have attempted to introduce technologies such as visible and near-infrared spectroscopy, laser speckle, and spatial frequency domain imaging into bedside monitoring, hoping to indirectly reflect microcirculatory disorders and inflammatory responses through real-time changes in tissue absorption, scattering, and hemodynamic parameters. However, existing optical solutions are still limited by multiple technical bottlenecks. First, the sparse wavelength range results in a low spectral information dimension, only able to acquire macroscopic average signals at a few discrete wavelengths, failing to resolve the fine spatiotemporal distribution of key molecules such as oxyhemoglobin, deoxyhemoglobin, and cytochrome in the microcirculatory network, thus weakening the ability to capture early, minute changes. Second, the inherent trade-off between spatial resolution and imaging depth makes it difficult for the system to simultaneously identify micron-level vascular structures and penetrate several millimeters of tissue within a centimeter-level field of view, limiting its applicability to superficial sites such as the skin and sublingual mucosa, and preventing effective detection of deep organs. Secondly, the size and optomechanical architecture of the equipment are limited by high-power light sources, precision beam splitters, and high-sensitivity detectors. The entire unit often weighs tens of kilograms and requires fixed brackets or rails for installation. This results in poor mobility in the confined space and complex pipeline environment of the ICU (Intensive Care Unit), making it difficult to achieve continuous monitoring from multiple positions and angles. Finally, optical signals are easily affected by background noise such as ambient stray light, patient motion artifacts, skin pigmentation, and wound exudate. Existing algorithms are not adaptable enough to dynamic and non-uniform media, leading to both false positives and false negatives. Clinicians still need to rely on experience for secondary screening, failing to truly reduce their workload.

[0005] In summary, how to achieve highly reliable, low-latency, and easily deployable early screening for sepsis has become an urgent problem to be solved by existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a sepsis screening system and method based on hyperspectral imaging, which overcomes the deficiencies of existing early sepsis identification methods such as insufficient sensitivity, poor real-time performance, low equipment flexibility, and large computational load, and realizes early and accurate screening of sepsis in real-time medical scenarios such as ICU bedside.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0008] The first aspect of this invention provides a sepsis screening system based on hyperspectral imaging, comprising:

[0009] The light source module is used to emit illumination light to the dual-channel multimode fiber optic transmission module.

[0010] The dual-channel multimode fiber optic transmission module is used to transmit illumination light to the data acquisition module and receive the optical signals acquired by the data acquisition module. The acquired optical signals are then converted into hyperspectral signals and input into the hyperspectral scanning imaging module.

[0011] The data acquisition module is used to acquire light signals from the tissue surface in real time through the principle of reflection;

[0012] The hyperspectral push-broom imaging module is used to convert the hyperspectral signal from the tissue surface into raw cubic data and transmit it to the data processing module;

[0013] The data processing module is used to process the original cubic data to obtain sepsis identification results. First, the original cubic data is dimensionality reduced. Then, a two-branch reconstruction network based on content matching gating mechanism is used to learn and identify sepsis features to detect abnormal regions. Finally, residual technology is used to further enhance the abnormal regions.

[0014] Furthermore, the light source module uses a broadband halogen lamp as the light source.

[0015] Furthermore, in the dual-channel multimode fiber optic transmission module, illumination light and received optical signals are transmitted through independent fiber optic channels respectively.

[0016] Furthermore, the data acquisition module includes a miniature focusing lens group composed of aspherical lenses and a folded optical path.

[0017] Furthermore, in the data processing module, the Mamba selective state-space model is used to perform real-time and efficient joint modeling of the spectral sequence and spatial spectrum of the original cubic data, thereby achieving dimensionality reduction.

[0018] Furthermore, in the data processing module, the dual-branch reconstruction network based on the content matching gating mechanism includes content matching gating, anomaly feature branch, and background feature branch. The content matching gating is used to dynamically determine the pattern of the input features. When it is determined to be an anomaly pattern, the feature is input into the anomaly feature branch, and when it is determined to be a background pattern, the feature is input into the background feature branch.

[0019] Furthermore, the abnormal feature branch employs a non-local attention mechanism to reconstruct abnormal region features through cross-regional feature weighted aggregation, thereby enhancing the spectral abnormality patterns associated with sepsis.

[0020] Furthermore, the background feature branch is used to optimize local spatial and spectral correlations, and uses a self-attention mechanism to smooth background features and eliminate noise interference.

[0021] Furthermore, it also includes a display module, including a display, for showing the identification results of sepsis.

[0022] A second aspect of the present invention provides a sepsis screening method based on hyperspectral imaging, comprising the following steps:

[0023] Illumination light is emitted from a light source and transmitted to a dual-channel multimode fiber optic transmission module;

[0024] The illumination light is transmitted to the data acquisition module, and the light signal on the tissue surface is acquired in real time through the principle of reflection.

[0025] The system receives the optical signal collected by the data acquisition module, converts the collected optical signal into a hyperspectral signal, and then inputs it into the hyperspectral scanning imaging module.

[0026] The hyperspectral signal from the tissue surface is converted into raw cubic data and transmitted to the data processing module.

[0027] The original cubic data is processed to obtain sepsis identification results. First, the original cubic data is dimensionality reduced. Then, a two-branch reconstruction network based on content matching gating mechanism is used to learn and identify sepsis features to detect abnormal regions. Finally, residual techniques are used to further enhance the abnormal regions.

[0028] The above one or more technical solutions have the following beneficial effects:

[0029] This invention discloses a sepsis screening system and method based on hyperspectral imaging. This invention achieves a closed-loop advantage in hardware architecture, algorithm flow, and clinical adaptability, realizing a non-invasive early sepsis screening system of "high-fidelity acquisition—low-latency processing—high-reliability output." First, a lightweight and robust front-end optical path is constructed using a dual-channel multimode fiber and a miniature focusing lens assembly, completely separating illumination and signal to eliminate backlight interference at its source, ensuring consistent spectral data even in complex clinical scenarios. At the algorithm level, the proposed hyperspectral state-space anomaly detection method transforms the originally high-dimensional, highly redundant hyperspectral cube into a low-rank, interpretable state sequence through multi-scale sliding windows, convolution-serialization, and efficient Mamba-SSM modeling. Its linear complexity significantly reduces the computational load compared to traditional quadratic complexity models, allowing the entire inference process to be completed in real-time at the edge. The content-matching gating and dual-branch reconstruction mechanism first quickly compares local features with prior templates, automatically deciding whether to invoke the "abnormal" or "background" thinking mode. The abnormal branch uses non-local attention to capture subtle fluctuations in metabolism and oxygenation across regions, while the background branch smooths out texture noise in normal tissues through self-attention. Together, they significantly suppress false alarms. Finally, through a deconvolution-residual-diffusion link, spatial details are restored while introducing morphological constraints and lesion priors, making abnormal areas interpretable in spatial, spectral, and morphological dimensions. Doctors can identify suspicious lesions without secondary interpretation.

[0030] This invention provides an early warning during the window period when tissue oxygenation has just deviated but blood counts have not yet changed, thus gaining valuable time for intervention. Driven by both engineering and algorithmic innovation, this invention brings hyperspectral imaging from the laboratory to real-world clinical procedures, significantly improving the timeliness, accuracy, and accessibility of sepsis prevention and control, and has broad application value.

[0031] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a structural diagram of the sepsis screening system based on hyperspectral imaging in Embodiment 1 of the present invention. Detailed Implementation

[0034] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0035] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0037] Example 1:

[0038] Embodiment 1 of the present invention provides a sepsis screening system based on hyperspectral imaging, such as... Figure 1As shown, it includes a light source module, a dual-channel multimode fiber optic transmission module, a data acquisition module, a data processing module, and a display module.

[0039] First, a non-contact portable imaging system with continuous spectral coverage of 400–2500 nm is constructed to achieve detailed observation of tissue microcirculation and metabolic status. Specifically, it includes a light source module, a dual-channel multimode fiber optic transmission module, a data acquisition module, and a hyperspectral push-broom imaging module.

[0040] The light source module is used to emit illumination light to the dual-channel multimode fiber optic transmission module.

[0041] In one specific implementation, the light source module uses a broadband halogen lamp as the light source, capable of emitting a continuous spectrum of 400–2500 nm.

[0042] The dual-channel multimode fiber optic transmission module is used to transmit illumination light to the data acquisition module and receive the optical signals acquired by the data acquisition module. After converting the acquired optical signals into hyperspectral signals, the signals are input into the hyperspectral scanning imaging module.

[0043] In one specific implementation, the dual-channel multimode fiber optic transmission module transmits illumination light and receives acquired optical signals through independent fiber optic channels, avoiding backlight interference common in traditional single-fiber systems. Specifically, illumination light is transmitted to the data acquisition module via channel 1. The miniature focusing lens assembly in the data acquisition module uniformly illuminates the patient tissue, obtaining the reflected signal, which is then transmitted back to the dual-channel multimode fiber optic transmission module via channel 2. The dual-channel multimode fiber optic transmission module then continues to transmit the hyperspectral signal to the hyperspectral push-broom imaging module via channel 2.

[0044] The dual-channel multimode fiber optic transmission module automatically adjusts the light source power based on tissue reflectivity to ensure spectral data consistency and achieve dynamic light intensity matching. This improves the signal-to-noise ratio, completely eliminates crosstalk during acquisition and illumination, and is suitable for complex clinical environments. Furthermore, the fiber optic cable in the dual-channel multimode fiber optic transmission module is flexible, adaptable to different detection sites.

[0045] The data acquisition module is used to acquire light signals from the tissue surface in real time through the principle of reflection.

[0046] In one specific implementation, the data acquisition module includes a miniature focusing lens assembly composed of aspherical lenses and a folded optical path. The miniature focusing lens assembly uses an aspherical lens design to replace the traditional multi-element lens assembly, correcting aberrations while compressing the optical path volume. The folded optical path, using a 45° reflecting mirror, reduces the physical size of the hyperspectral pushbroom imaging module by 50%. This results in a lightweight design, supports handheld inspection, and enables rapid focusing.

[0047] The light signal reflected from the tissue surface is captured and focused by the aspherical lens group, then folded by the 45° reflector into the hyperspectral push-broom imaging module for dispersion and conversion into an electrical signal, generating hyperspectral raw cubic data, which provides input for subsequent data processing modules.

[0048] The hyperspectral push-broom imaging module is used to convert hyperspectral signals from the tissue surface into raw cubic data for transmission to the data processing module. After receiving the transmitted hyperspectral signals from the tissue surface, the module first uses an internal grating to decompose the broadband light into discrete monochromatic light according to wavelength. Then, a CMOS imaging sensor converts the monochromatic light signals into analog electrical signals, which are then converted into digital signals by an analog-to-digital converter. Simultaneously, the push-broom scanning mechanism scans and acquires spatial information along a one-dimensional direction. Finally, the digital signals of different wavelengths are integrated according to spatial (X, Y dimensions) and spectral (λ dimension) dimensions into three-dimensional raw cubic data and transmitted to the data processing module.

[0049] The data processing module is used to process the raw cube data to obtain sepsis identification results.

[0050] In one specific implementation, the data processing module introduces the Mamba Selective State Space Model (Mamba-SSM) for real-time and efficient joint spatial-spectral modeling of spectral sequences. It combines the spatial and spectral dimensions of the data for modeling, processing spectral information in sequence and associating spatial pixel relationships. This comprehensively captures spatial structural differences and spectral feature variations on the tissue surface, reducing computational complexity to O(NL) and significantly improving inference speed. A dual-branch reconstruction network based on Content Matching Gating (CMM) is designed to effectively reduce the false detection rate of anomalies. Residual diffusion (RD) post-processing technology is employed to further enhance the contrast between the anomaly region and the background, improving detection accuracy. Specifically, the original cubic data is first dimensionality-reduced, then a dual-branch reconstruction network based on Content Matching Gating is used for sepsis feature learning and recognition to detect anomaly regions. Finally, residual techniques are used to further enhance the anomaly regions.

[0051] More specifically, it includes the following steps:

[0052] Step 1: Preprocess the original cube data.

[0053] This embodiment utilizes a hyperspectral imaging device to acquire raw cubic data of the tissue surface: . Represents the real number field. Corresponding to the spatial height dimension, Corresponding to the spatial width dimension, Corresponding to the spectral dimension. Raw cube data is a three-dimensional data structure containing two spatial dimensions (X and Y directions) and one spectral dimension λ. The spatial dimensions represent the image pixel locations. Raw cube data can record reflectance or absorptivity information across continuous narrow bands.

[0054] Background noise and equipment non-uniformity are eliminated through black and white point correction: .

[0055] in, It is the corrected original cube data, which is the result obtained after dark field and white field correction processing, and is used for subsequent analysis or application. This is the raw cube data, which is the initial input data before any calibration operation. This is for dark reference data. For reference only. It is an extremely small positive number (epsilon), and its main function is to avoid the case where the denominator is 0, thus ensuring the stability of the formula during numerical calculation.

[0056] Step 2: Perform sliding window segmentation, convolutional coding, and serialization on the preprocessed original cube data.

[0057] This embodiment uses a 9×9 pixel window to slide and segment the image with a 3-pixel stride to generate a spatial sub-block sequence: .

[0058] in Hyperspectral data representing the i-th spatial sub-block This represents the total number of spatial sub-blocks obtained after sliding partitioning. Then, a 3×3 convolution is applied to each sub-block to map it into a feature map, which is then flattened into a one-dimensional sequence for use in the state-space model.

[0059] .

[0060] in, It is the first Each space sub-block The feature tensor output after 3×3 convolution (dimension) ); This indicates the number of feature channels generated by the convolution operation; It is The 9x9 spatial dimension is flattened as Feature sequence after dimension (81×) ).

[0061] Step 3: Utilize the Mamba selective state-space model to perform real-time and efficient joint modeling of the spectral sequence and spatial spectrum of the original cubic data, thereby achieving dimensionality reduction.

[0062] This embodiment uses the standard equation of state. Improved to:

[0063] .

[0064] in, yes The hidden state at all times It is the state transition matrix. For spectral dimensions The state transition matrix, yes The hidden state at all times It is the input weight matrix. yes Input at any time It is a mask vector corresponding to dimensions 320–580.

[0065] Incorporating tissue oxygenation features into the standard discrete convolution formula:

[0066] .

[0067] in, It is a discrete convolutional output that integrates tissue oxygenation features. for From 0 to The summation operation, where L is the number of convolution kernels. It is the first Each convolutional kernel parameter, It is the input signal. Indicates downsampling Each step size corresponds to a convolution kernel parameter. It is a complementary and integrated weight. These are parameters associated with tissue oxygenation features. α∈[0.4,0.8] represents adaptive weights based on local blood oxygen saturation. The complexity is O(NL), significantly lower than the Transformer's O(N²) complexity, enabling real-time operation.

[0068] Step 4: The dual-branch reconstruction network based on the content matching gating mechanism includes the content matching gate (CMG), the anomaly feature branch (AFB), and the background feature branch (BFB).

[0069] The content matching gating mechanism dynamically determines the pattern of input features. It calculates the similarity between the current sub-block features and a preset template library, compares this similarity to a dynamic threshold, and determines the subsequent processing path. If the matching score is below the threshold, it is identified as an abnormal pattern; otherwise, it is identified as a background pattern. When an abnormal pattern is identified, the features are input into the abnormal feature branch; when a background pattern is identified, the features are input into the background feature branch. This mechanism effectively reduces false detections and improves the robustness of the algorithm in complex contexts.

[0070] The abnormal feature branch employs a non-local attention mechanism, reconstructing abnormal region features through cross-regional feature weighting and aggregation, thereby enhancing spectral abnormality patterns associated with sepsis. Its output is the reconstructed abnormal region features. This branch is particularly suitable for detecting local metabolic abnormalities or changes in oxygenation, improving the sensitivity of early sepsis identification.

[0071] The background feature branch optimizes local spatial and spectral correlations, using a self-attention mechanism to smooth background features and eliminate noise interference. Its output is the optimized normal region (background) feature. This branch ensures low response in normal tissue areas, avoiding misclassification as abnormalities and thus reducing the false positive rate.

[0072] More specifically:

[0073] Define a content matching threshold τc = 0.85 to determine whether the input feature is an anomalous pattern (match score below this value) or a background pattern (match score above this value), and dynamically select the anomalous feature branch (AFB) or background feature branch (BFB) for feature reconstruction:

[0074] AFB refactoring method for abnormal branches:

[0075] .

[0076] in, It is a node The reconstruction features, right Neighborhood set Summation, yes with neighbors The weight, For normalization function, It is a node The original characteristics, These are the original characteristics of the neighbors. It is the characteristic transformation matrix.

[0077] Background branch BFB enhances local spatial spectral correlation:

[0078] .

[0079] in, It is an output that enhances the correlation of local spatial spectra. It is a normalization function, where Q is the query matrix and K is the key matrix. Q is the transpose of K, d is the dimension of Q and K (used for scaling), and V is the value matrix.

[0080] Step 5: Deconvolution feature reconstruction.

[0081] Deconvolutional feature reconstruction targets the "abnormal region features" output by the Anomalous Feature Branch (AFB) and the "normal region features" output by the Background Feature Branch (BFB) in the previous step, performing deconvolution reconstruction on each separately. A 3×3 convolutional transpose network is then used to restore the processed feature sequence to its original spatial scale. .

[0082] in, It is to restore the feature sequence at the original spatial scale. This represents a 3×3 transpose convolution operation. This is the feature sequence before processing by the convolutional transpose network.

[0083] A 3×3 convolutional transpose network was used to restore the processed feature sequence to its original spatial scale, and optimizations were made for sepsis detection: a post-processing module based on the morphology of sepsis lesions was added after the deconvolution layer, and the reconstruction results were corrected by analyzing the geometric features of the lesions (circularity ≥ 0.7, boundary clarity ≥ 0.8); the deconvolution kernel parameters were optimized based on clinical data, with a focus on enhancing the reconstruction accuracy of the 520-580nm feature band.

[0084] Step 6: Use residual diffusion post-processing technology to further enhance the contrast between abnormal areas and the background.

[0085] First, residual calculation is performed, and the difference between the original image and the reconstructed image is used for anomaly detection: .

[0086] in, It is the residual feature at position i, which quantifies the difference between the original and reconstructed features and is a core indicator for anomaly detection. The original image features at position i, It is the reconstructed feature at position i, the model's fitting result to the normal pattern, and then residual diffusion is performed.

[0087] Employing lesion morphology-adaptive three-dimensional pooling kernels to highlight the spatial continuity of abnormal regions:

[0088] .

[0089] in, It is a pooling output feature. The representative core size is Three-dimensional average pooling operation, These are the residual features of the input. Vascular region: 3×3×3 nuclei, subcutaneous tissue: 5×5×5 nuclei, boundary transition region: 3×5×3 nuclei.

[0090] Step 7: Generate the final suspected infection area.

[0091] First, heatmap generation and normalization are performed. The residual diffusion results of all sub-blocks are aggregated and normalized to the 0–1 range to visually represent the risk distribution.

[0092] .

[0093] in, It is the target matrix. Represented as (High) × The (wide) real matrix needs to have its values ​​normalized to the [0,1] interval.

[0094] Next, binarization thresholding is performed. The Otsu method is used to adaptively determine the threshold, obtaining the final mask output of the suspected infected area.

[0095] .

[0096] in, It is a location The binary output at the location is 1. This is an indicator function (it takes the value 1 if the condition is met, otherwise 0). It is the target matrix exist The element at that location, It is the classification threshold.

[0097] The display module includes a monitor, and the monitor's medical display interface is used to display the identification results of sepsis.

[0098] Example 2:

[0099] Embodiment 2 of the present invention provides a sepsis screening method based on hyperspectral imaging, comprising the following steps:

[0100] Illumination light is emitted from a light source and transmitted to a dual-channel multimode fiber optic transmission module;

[0101] The illumination light is transmitted to the data acquisition module, and the light signal on the tissue surface is acquired in real time through the principle of reflection.

[0102] The system receives the optical signal collected by the data acquisition module, converts the collected optical signal into a hyperspectral signal, and then inputs it into the hyperspectral scanning imaging module.

[0103] The hyperspectral signal from the tissue surface is converted into raw cubic data and transmitted to the data processing module.

[0104] The original cubic data is processed to obtain sepsis identification results. First, the original cubic data is dimensionality reduced. Then, a two-branch reconstruction network based on content matching gating mechanism is used to learn and identify sepsis features to detect abnormal regions. Finally, residual techniques are used to further enhance the abnormal regions.

[0105] The steps and methods involved in the above embodiment two correspond to those in embodiment one. For specific implementation details, please refer to the relevant description section of embodiment one.

[0106] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0107] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0108] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A hyperspectral imaging based sepsis screening system, characterized in that, The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. ； ； in, yes The hidden state at all times It is the state transition matrix. For spectral dimensions The state transition matrix, yes The hidden state at all times It is the input weight matrix. yes Input at any time It is a mask vector corresponding to dimensions 520–580; It is a discrete convolutional output that integrates tissue oxygenation features. for From 0 to The summation operation, where L is the number of convolution kernels. It is the first Each convolutional kernel parameter, It is the input signal. Indicates downsampling Each step size corresponds to a convolutional kernel parameter. It is a complementary and integrated weight. It is a parameter associated with tissue oxygenation characteristics. The adaptive weights are based on local blood oxygen saturation. The application relates to a sepsis recognition method and device.

2. The hyperspectral imaging-based sepsis screening system of claim 1, wherein, The application relates to a sepsis recognition method and device.

3. The hyperspectral imaging-based sepsis screening system as claimed in claim 1, wherein, The application relates to a sepsis recognition method and device.

4. The hyperspectral imaging-based sepsis screening system of claim 1, wherein, The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a sepsis recognition method and device. The application relates to a seps

Images