Blood oxygen distribution imaging method and system based on dual-band diffuse reflection space mapping

Abstract

The invention discloses a blood oxygen distribution imaging method and system based on dual-band diffuse reflection space mapping, and relates to the field of non-contact physiological parameter detection for biomedical optical application. The method comprises the steps that a dual-band diffuse reflection original image sequence of a to-be-detected tissue area under two specific bands is obtained, performing preprocessing and region segmentation to obtain a normalized dual-band image and a region of interest; estimating an effective optical path ratio between two specific wavebands based on image gradient information in a local neighborhood of each pixel, and constructing a local optical path correction matrix; according to the dual-band reflection attenuation ratio, the corresponding effective optical path ratio and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in two specific bands, a dual-band inversion model based on the improved Lambert-Beer law is used for solving initial blood oxygen saturation, and a two-dimensional blood oxygen saturation distribution diagram is generated through spatial regularization optimization. Therefore, the spatial characterization capability, the inversion stability and the imaging accuracy of non-contact blood oxygen imaging can be improved.

Description

Technical Field

[0001] This application relates to the field of non-contact physiological parameter detection for biomedical optical applications, and in particular to a blood oxygen distribution imaging method and system based on dual-band diffuse reflection spatial mapping. Background Technology

[0002] Blood oxygen saturation is an important parameter reflecting the oxygen supply status and physiological function of human tissues, and it plays a significant role in health monitoring, critical care, skin lesion assessment, and local tissue blood supply analysis. Currently, blood oxygen detection typically relies on contact measurement devices, which acquire blood oxygen information from local areas of the body through light transmission or reflection. While this method is relatively mature, it usually requires close contact with the skin, which can easily cause local pressure and discomfort. Furthermore, its use is often limited in special application scenarios such as burn care, skin abnormalities, and infection control.

[0003] With the development of contactless physiological parameter detection technology, long-distance blood oxygenation detection of target areas using optical imaging methods has gradually attracted attention. Inferring blood oxygenation status by collecting reflected light signals from the human body surface offers advantages in reducing contact risks and improving ease of use. However, from the current state of technological development, non-contact blood oxygenation detection still faces significant limitations in practical applications: one approach relies on complex imaging equipment such as multispectral or hyperspectral sensors, which, while capable of acquiring rich spectral information, generally suffer from complex equipment, high costs, low data acquisition efficiency, and cumbersome calibration processes; another approach is based on simplified implementations using ordinary camera equipment, which, while having a relatively lower hardware threshold, often struggles to meet the high-precision application requirements in terms of signal quality, anti-interference capabilities, and measurement stability.

[0004] Furthermore, current non-contact blood oxygenation detection technologies still have limited ability to characterize spatial differences in blood oxygenation status within a target area, and there is still room for improvement in measurement stability and imaging accuracy under the influence of complex tissue optical properties. Summary of the Invention

[0005] This application provides a blood oxygen distribution imaging method, system, storage medium, computer program product, and electronic device based on dual-band diffuse reflection spatial mapping, which at least solves the problems of insufficient measurement accuracy and spatial distribution characterization ability in the prior art of non-contact blood oxygen detection.

[0006] In a first aspect, embodiments of this application provide a blood oxygen distribution imaging method based on dual-band diffuse reflectance spatial mapping. The method includes: acquiring a sequence of original dual-band diffuse reflectance images of a tissue region under two specific bands; preprocessing and segmenting the original dual-band diffuse reflectance image sequence to obtain a normalized dual-band image and a corresponding region of interest; for each pixel within the region of interest, estimating the effective optical path ratio between the two specific bands based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, to construct a local optical path correction matrix covering the region of interest; and for each pixel within the region of interest, adjusting the normalized dual-band image according to the image gradient information of the local neighborhood of the corresponding pixel. The initial oxygen saturation of a pixel is calculated using a preset dual-band inversion model, based on the reflection attenuation ratio in the two specific bands and the effective optical path ratio of the corresponding pixel in the local optical path correction matrix. The dual-band inversion model is configured to: establish a functional mapping relationship between the reflection attenuation ratio, the effective optical path ratio, the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the two specific bands, and the initial oxygen saturation, based on the improved Lambert-Beer law, to obtain the initial oxygen saturation of the corresponding pixel; and perform spatial regularization optimization on the initial oxygen saturation of all pixels in the region of interest to generate a two-dimensional oxygen saturation distribution map corresponding to the region of interest.

[0007] Secondly, embodiments of this application provide a blood oxygen distribution imaging system based on dual-band diffuse reflectance spatial mapping. The system includes: a dual-band image acquisition unit, used to acquire a sequence of original dual-band diffuse reflectance images of a tissue region under two specific bands; an image processing unit, used to preprocess and segment the original dual-band diffuse reflectance image sequence to obtain a normalized dual-band image and a corresponding region of interest; an optical path correction unit, used to estimate the effective optical path ratio between the two specific bands for each pixel within the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, to construct a local optical path correction matrix covering the region of interest; and a blood oxygen inversion unit, used to preprocess and segment the original dual-band diffuse reflectance image of a tissue region under two specific bands. For each pixel, based on the reflection attenuation ratio of the normalized dual-band image under the two specific bands, and combined with the effective optical path ratio of the corresponding pixel in the local optical path correction matrix, the preliminary blood oxygen saturation of the pixel is calculated using a preset dual-band inversion model. The dual-band inversion model is configured to: establish a functional mapping relationship between the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin under the two specific bands and the preliminary blood oxygen saturation based on the improved Lambert-Beer law, so as to solve for the preliminary blood oxygen saturation of the corresponding pixel; a distribution imaging unit is used to perform spatial regularization optimization on the preliminary blood oxygen saturation of all pixels in the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping of any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to any embodiment of this application.

[0011] The blood oxygen distribution imaging method and system based on dual-band diffuse reflection spatial mapping provided in this application can achieve at least the following technical effects:

[0012] (1) By combining the image gradient information of the dual-band normalized image in the local neighborhood of each pixel, the effective optical path ratio of the corresponding pixel between the two specific bands is estimated, and a local optical path correction matrix covering the region of interest is further constructed. Based on this processing method, the ideal assumption of treating the target tissue as a uniform optical medium in traditional non-contact measurement is broken. It enables the dual-band diffuse reflection information to be adapted to the local texture changes and microstructural differences of the tissue surface before entering the subsequent blood oxygen solution, thereby achieving targeted correction of optical path distortion caused by non-uniform scattering at the pixel level. As a result, not only is the consistency and comparability of reflection information at different spatial locations significantly enhanced, but a reliable data foundation is also provided for subsequent high-fidelity blood oxygen parameter inversion.

[0013] (2) The dual-band reflection attenuation ratio is coupled with the effective optical path ratio of the corresponding pixel, and a dual-band inversion model is constructed based on the improved Lambert-Beer law to incorporate the extinction characteristics of oxyhemoglobin and deoxyhemoglobin in two specific bands into the blood oxygen saturation solution process. On this basis, the spatial regularization optimization of the preliminary blood oxygen saturation results is further combined. Through the above coupling design, on the one hand, the blood oxygen solution process is given a strict underlying physical mechanism constraint, which effectively decouples the tissue oxygenation state; on the other hand, the spatial regularization mechanism effectively suppresses the local noise disturbance and isolated artifacts that are easily amplified in pixel-level independent inversion, so that the final generated two-dimensional blood oxygen saturation distribution map is smoother, more coherent and more interpretable while maintaining the true trend of blood oxygen change in the tissue area, which is more conducive to the fine imaging characterization of the target tissue area.

[0014] This technical solution constructs an integrated processing link around the dual-band diffuse reflection imaging process, encompassing "local optical path difference correction—physical mechanism decoupling and inversion—spatial distribution coherent optimization." This not only effectively overcomes the optical path deviation defects caused by non-uniform scattering at the physical level but also significantly improves noise resistance through rigorous algorithmic spatial constraints. Consequently, non-contact pulse oximetry detection is upgraded from coarse regional mean estimation to high-resolution two-dimensional distribution imaging for complex physiological tissues, significantly enhancing the visualization capabilities and reliability of pulse oximetry results in clinical diagnosis and physiological monitoring. Attached Figure Description

[0015] 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an example of a blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown;

[0017] Figure 2 A flowchart illustrating an example of obtaining a normalized dual-band image and the corresponding region of interest using a method according to an embodiment of this application is shown.

[0018] Figure 3 A flowchart illustrating an example of calculating the preliminary blood oxygen saturation of a pixel using a dual-band inversion model according to an embodiment of this application is shown.

[0019] Figure 4 A schematic diagram illustrating the system operation mechanism of an example blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown.

[0020] Figure 5 A schematic diagram of a comparative experimental simulation is shown, illustrating an example of imaging in complex microvascular networks and localized ischemic areas.

[0021] Figure 6 A schematic diagram illustrating the simulation effect of a comparative experiment under individual differences in different skin colors is shown.

[0022] Figure 7 A schematic diagram of the comparative experimental simulation results is shown as an example in the discussion of algorithmic trade-off analysis and the limits of microscopic spatial resolution.

[0023] Figure 8 A structural block diagram of an example of a blood oxygen distribution imaging system based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be noted that in blood oxygenation detection based on the principle of optical absorption, the propagation of light signals of different wavelengths within biological tissues is simultaneously affected by both absorption and scattering. The inversion result of blood oxygen saturation is usually closely related to the difference in effective optical path length at different wavelengths. Currently, contact-based dual-wavelength pulse oximetry schemes, such as the CFPO (Calibration-Free Pulse Oximetry) approach, typically rely on empirical calibration relationships to establish the correspondence between dual-wavelength absorption information and blood oxygen saturation. Although some schemes have attempted to reduce calibration dependence by selecting adjacent bands and assuming an approximately constant optical path ratio, these methods are usually limited to single-point measurements and lack effective compensation for local optical path length differences caused by variations in skin color, tissue thickness, and local tissue structure between individuals. Therefore, it is difficult to simultaneously meet the requirements of measurement accuracy and spatial distribution characterization.

[0026] In the field of non-contact blood oxygenation detection, some existing technologies utilize multi-band image acquisition and statistical modeling to infer blood oxygen levels. For example, the DS-based non-contact SpO2 (Dynamic Spectrum-based non-contact oxygen saturation detection) approach acquires video sequences at multiple wavelengths, extracts pulsation-related spectral features, and combines them with a regression model to output blood oxygenation results. The DR-HSI (Diffuse Reflectance-Hyperspectral Imaging) approach attempts to combine diffuse reflectance imaging with high-dimensional spectral imaging to simultaneously utilize spatial image information and spectral information to achieve local blood oxygenation assessment. These methods can improve the information dimension to some extent and have the basic conditions for acquiring regional images. However, they usually rely on a lot of narrow-band data acquisition, the system is highly complex, and the subsequent processing is more based on global statistical feature extraction, empirical modeling or basic segmentation processing. They fail to fully consider the influence of tissue thickness variation and spatial non-uniform scattering on the local light propagation path. Therefore, it is still difficult to provide a stable and precise quantitative characterization of the true oxygenation state at each pixel location.

[0027] On the other hand, to reduce hardware complexity, some related technologies currently attempt to use ordinary camera equipment or simplified dual-band illumination structures for blood oxygen assessment. For example, the RGB-SpO2 (RGB camera-based oxygensaturation measurement) approach uses improved ordinary camera equipment for non-contact blood oxygen estimation, while the SUDWI pulse oximetry imaging (Spatially Uniform DualWavelength Illumination pulse oximetry imaging) approach improves spatial illumination consistency by constructing a dual-band uniform illumination system.

[0028] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0029] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0030] Figure 1 A flowchart illustrating an example of a blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown.

[0031] Regarding the execution subject of the method in this application embodiment, it can be any controller or processor with computing or processing capabilities, such as an optical imaging processing controller. It implements the method of this application embodiment by running programs or instructions stored in a storage medium. In practical business scenarios, this method can be applied to non-contact vital sign monitoring in intensive care units (ICUs), assessment of skin flap blood supply in patients with large-area burns or trauma, non-contact health monitoring of newborns, and dynamic observation of localized ischemic areas of the skin. Unlike traditional contact-based single-point blood oxygenation measurement methods, this embodiment can perform two-dimensional spatial imaging of blood oxygenation status in a large area of ​​tissue without contacting the object being measured.

[0032] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0033] like Figure 1As shown, in step S110, a sequence of original dual-band diffuse reflection images of the tissue region to be tested in two specific bands is obtained.

[0034] Specifically, since oxyhemoglobin and deoxyhemoglobin have different absorption response characteristics at different optical bands, this embodiment provides two specific bands of probe light to the tissue area under test and uses an imaging device with corresponding band response capabilities to acquire the backscattered optical signal of the tissue area under test. The imaging device can be an area array camera with a corresponding filter structure, a multi-channel imaging module, or other sensors capable of acquiring images at two specific bands. In this way, raw image data reflecting the difference in optical response of the tissue area under test under dual-band conditions can be obtained. For example, in an ICU ward, the imaging device can be positioned at a certain distance above the patient's forearm or back of the hand to continuously acquire dual-band images without contacting the patient's skin, thus avoiding additional pressure on the patient's local tissues caused by catheters, dressing materials, or contact probes.

[0035] In some implementations, to preserve the dynamic information of the tissue region under test changing with the cardiac cycle, the system can continuously acquire image sequences containing a time dimension. These raw image sequences not only record the spatial distribution of reflected light intensity on the target area's surface but also include temporal variations caused by cyclical changes in blood volume. For example, in the scenario of observing blood supply after flap transplantation, continuously acquired dual-band image sequences can reflect both the overall surface reflection changes of the flap and preserve the dynamic information of its microcirculation state over time. This provides a raw data foundation for subsequent assessments of whether local blood supply is uniform and whether early hypoperfusion areas have appeared, enabling the system to acquire basic optical data covering the area under test in a non-contact manner.

[0036] In step S120, the original dual-band diffuse reflection image sequence is preprocessed and segmented to obtain a normalized dual-band image and its corresponding region of interest.

[0037] It should be noted that in the actual imaging process, the original image sequence is usually affected by factors such as ambient background light, uneven illumination, sensor dark response, and slight movement of the object being measured. If the original image is used directly for subsequent inversion, it is easy to cause a lack of a unified benchmark for light intensity values ​​at different locations, which in turn affects the comparability of dual-band information.

[0038] Therefore, the steps in this embodiment first perform preprocessing on the original dual-band diffuse reflectance image sequence to correct and suppress non-ideal acquisition factors in the original image, thereby obtaining a normalized dual-band image with uniform scale and high comparability. For example, in a neonatal non-intrusive monitoring scenario, the monitored subject may exhibit slight limb movements or overall displacement driven by breathing. If spatiotemporal alignment and intensity normalization are not performed, the dual-band reflectance information between different time frames is prone to mismatch, affecting the subsequent accurate analysis of blood oxygenation status.

[0039] After basic preprocessing, this embodiment further performs region segmentation to extract effective tissue regions directly related to blood oxygenation analysis from the overall field of view. Specifically, irrelevant areas such as clothing, hair, medical accessories, and non-physiological backgrounds can be excluded by combining the image's reflection features, texture features, edge features, or other image attributes, while retaining regions of interest containing skin tissue and effective physiological structures. For example, in burn care or trauma monitoring scenarios, the area to be tested may be surrounded by dressing edges, fixation tape, support pads, or medical tubing. Without prior region segmentation, the high reflectivity or abnormal textures of these non-tissue areas will interfere with subsequent local optical path estimation and blood oxygenation inversion. Through the synergistic effect of preprocessing and region segmentation, the stability and signal-to-noise ratio of the input data can be improved, and subsequent calculations can be focused on physiologically significant target areas.

[0040] In step S130, for each pixel in the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, the effective optical path ratio between the two specific bands of each pixel is estimated to construct a local optical path correction matrix covering the region of interest.

[0041] In real biological tissues, the propagation path of photons within the tissue typically exhibits significant spatial variations due to differences in skin color, tissue thickness, and the heterogeneity of microstructure. If the assumption of a globally constant optical path, treating the entire target region as a homogeneous medium, is still adopted, it can easily introduce substantial inversion biases in two-dimensional imaging scenarios.

[0042] Therefore, this embodiment no longer uses a globally uniform optical path ratio. Instead, it utilizes image gradient information within the local neighborhood of the target pixel to characterize the local reflection attenuation changes near that location, thereby further inferring the local propagation differences under dual-band conditions. For example, in the edge region of a skin ulcer, normal and abnormal tissues often differ in thickness, water content, and microcirculation conditions, and the light propagation paths of adjacent regions may be significantly different. If a uniform optical path ratio is used for processing in this case, it can easily lead to distortion in the blood oxygen estimation of the edge region.

[0043] Specifically, this embodiment analyzes the spatial variation trend of the normalized dual-band image within the neighborhood of each pixel in the region of interest (ROI), and estimates the effective optical path ratio (APR) at each location pixel by pixel through the correspondence between the local gradient distributions under the two specific bands. Subsequently, the APRs corresponding to all pixels in the ROI are spatially organized to form a local optical path correction matrix covering the entire ROI. This local optical path correction matrix reflects the changes in light propagation paths caused by local scattering differences within the tested tissue region, thus providing pixel-level optical path compensation for subsequent dual-band inversion. For example, in the periphery of wounds in patients with large-area burns, the tissue surface state and subcutaneous structure are often highly heterogeneous. By constructing a local optical path correction matrix, subsequent inversion can more closely approximate the real optical conditions at each local location, rather than relying on a single empirical constant. Therefore, the influence of tissue non-uniform optical properties on blood oxygen estimation is incorporated into the local correction process, improving the adaptability of subsequent inversion results to complex tissue conditions.

[0044] In step S140, for each pixel in the region of interest, the preliminary blood oxygen saturation of the pixel is calculated using a preset dual-band inversion model, based on the reflection attenuation ratio of the normalized dual-band image in two specific bands and the effective optical path ratio of the corresponding pixel in the local optical path correction matrix.

[0045] Here, the dual-band inversion model is configured as follows: based on the improved Lambert-Beer law, a functional mapping relationship is established between the reflection attenuation ratio, effective optical path ratio, extinction coefficients of oxyhemoglobin and deoxyhemoglobin in two specific bands and the preliminary blood oxygen saturation, so as to solve for the preliminary blood oxygen saturation of the corresponding pixel.

[0046] In some implementations, the reflection attenuation ratio at each pixel of the normalized dual-band image can be extracted first, serving as an observation characterizing the differences in tissue absorption across different bands. Subsequently, this observation, the effective optical path ratio corresponding to that pixel, and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the two specific bands are substituted into the dual-band inversion model. Since this model is based on a modified Lambert-Beer law, it can establish a clear mapping relationship between dual-band optical observations and hemoglobin oxygenation status under local optical path correction, and thus solve for the preliminary blood oxygen saturation at each pixel location. For example, in a flap blood supply assessment scenario, if a local area experiences a decrease in the proportion of oxyhemoglobin due to insufficient blood supply, the reflection attenuation relationship of that area under dual bands will change accordingly. This embodiment can combine local optical path correction information to perform a physical inversion of this change, thereby obtaining pixel-level results that more closely approximate the true local oxygenation state.

[0047] This embodiment achieves point-by-point inversion of each pixel within the region of interest, rather than being limited to outputting the average blood oxygenation value of the entire region. Simultaneously, because the effective optical path ratio of the corresponding pixel in the local optical path correction matrix has been incorporated into the inversion process, the dual-band observation information at different pixel locations can be interpreted under conditions more consistent with the optical characteristics of local tissues, thereby improving the physical consistency and spatial resolution of the preliminary blood oxygenation results. For example, in the boundary region between a local ischemic lesion and surrounding normal tissue, a pixel-level oxygenation distribution with spatial transition relationships can be initially formed, rather than obtaining only a single, difficult-to-locate overall value.

[0048] In step S150, spatial regularization optimization is performed on the preliminary blood oxygen saturation of all pixels within the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

[0049] It should be noted that since the blood oxygen calculation in step S140 is performed independently for each pixel, the preliminary blood oxygen saturation result may still be affected by factors such as local noise, residual motion disturbances and single-point estimation errors, resulting in isolated noise points, abnormal jumps or spatial discontinuities in local locations.

[0050] Considering that blood oxygen distribution in biological tissues typically exhibits a certain degree of spatial consistency, while real local variations may exist at structural boundaries, this embodiment further performs spatial regularization optimization on the initial blood oxygen saturation to preserve as much realistic spatial boundary information as possible while suppressing random noise. For example, at the edges of burn wounds or ischemic skin areas, there may be significant local oxygenation changes, which are also prone to noise perturbation. Without spatial optimization, artifacts and unstable local jumps may appear simultaneously in the final image.

[0051] Specifically, in this embodiment, the initial blood oxygen saturation of each pixel within the region of interest is used as the initial optimization value. While maintaining the overall trend of the original inversion results, appropriate spatial constraints are applied to the blood oxygen differences between adjacent pixels, causing the blood oxygen distribution to gradually converge towards a result that better conforms to the spatial continuity of tissue. This effectively reduces the impact of isolated local artifacts and random perturbations on the final image quality, ultimately generating a two-dimensional blood oxygen saturation distribution map that corresponds one-to-one with the spatial location of the region of interest. For example, in non-contact monitoring of large skin areas in ICU patients, the results optimized by spatial regularization can more clearly present the spatial distribution characteristics of areas with local hypoperfusion, abnormal blood supply, or lesion edges, thus providing medical personnel with more intuitive, stable, and valuable imaging results.

[0052] In some examples of embodiments of this application, before step S110, the center wavelengths corresponding to two specific bands can be determined based on an adaptive wavelength selection mechanism. This optimized implementation overcomes the limitations of fixed illumination wavelength configuration in traditional blood oxygenation detection, allowing the selected wavelength combination to better adapt to differences in skin color, tissue thickness, and local tissue conditions among different subjects.

[0053] More specifically, firstly, the initial skin color features and tissue thickness features corresponding to the tissue region to be tested are extracted and converted into the corresponding melanin index and tissue thickness parameters, respectively.

[0054] In practice, the system can acquire initial reflectance information of the tissue region under test through low-power broadband pre-scanning, or it can use a visible light imaging device to acquire a reference image. Combined with color space conversion, regional morphology analysis, or contour scale analysis, the system extracts the skin color and tissue shape features of the tissue region under test. Subsequently, the extracted basic features can be converted into a quantified melanin index through pre-calibrated mapping relationships, lookup tables, or parameter fitting models. With tissue thickness parameters Therefore, individualized physiological characterization parameters corresponding to the current subject can be established before wavelength selection, thus providing boundary conditions for the sensitivity evaluation of subsequent candidate wavelengths.

[0055] Then, in the preset candidate wavelength spectral library, the sensitivity factor corresponding to each candidate wavelength is calculated by combining the extinction coefficients of oxyhemoglobin and deoxyhemoglobin. The sensitivity factor is proportional to the absolute value of the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin at the same candidate wavelength, and inversely proportional to the standard deviation of system noise, which is jointly affected by the melanin index and tissue thickness parameters.

[0056] For example, the sensitivity factor of the candidate wavelength It can be obtained by calculation using the following formula:

[0057] Equation (1)

[0058] In the formula, and These represent oxyhemoglobin and deoxyhemoglobin at candidate wavelengths, respectively. The extinction coefficient at that point, Indicates at candidate wavelength The following is based on the melanin index and tissue thickness parameters The standard deviation of system noise due to common influences.

[0059] In some embodiments, in order to more fully disclose the modulation mechanism of the system noise standard deviation in the specification, the aforementioned system noise standard deviation... Further quantitative calculations can be performed using the following model:

[0060] Equation (2)

[0061] In the formula, Indicates the optical system at wavelength Baseline equipment noise parameters at the location; For melanin at wavelength The equivalent absorption modulation coefficient at the location; For physiological tissues at wavelength The equivalent scattering attenuation coefficient at that location.

[0062] As can be seen from the above relationship, the numerator of the sensitivity factor reflects the ability of the candidate wavelength to distinguish differences in hemoglobin oxygenation status, while the denominator characterizes the system noise level corresponding to that candidate wavelength under the current individual conditions. With increasing melanin index and tissue thickness parameters, the absorption and scattering effects of light in the tissue generally increase, thereby weakening the effective backscattered signal and affecting the system's signal-to-noise performance at the corresponding wavelength. Therefore, the system can quantitatively evaluate the effective detection capability of different candidate wavelengths under individualized conditions, thus providing a basis for subsequent dual-band combination selection.

[0063] Then, iterate through all candidate wavelength combinations in the candidate wavelength spectral library and construct an optimization function with the goal of maximizing the sum of the sensitivity factors corresponding to two candidate wavelengths and with the spectral distance between the two candidate wavelengths satisfying the preset constraint condition.

[0064] For example, the optimization function can be expressed as follows:

[0065] Equation (3)

[0066] In the formula, and These represent the two candidate wavelengths that constitute the wavelength combination to be evaluated. This is a preset minimum spectral distance threshold used to avoid excessively small local optical path ratios. This is a preset maximum spectral distance threshold used to avoid excessive differences in tissue scattering.

[0067] Then, the optimization function is solved to obtain the optimal wavelength combination, and the two candidate wavelengths corresponding to the optimal wavelength combination are... and The center wavelengths corresponding to two specific wavebands are determined; the preset constraints include that the spectral distance between the two candidate wavelengths is greater than or equal to the minimum spectral distance threshold used to avoid excessively small local optical path ratios. And less than or equal to the maximum spectral distance threshold used to avoid excessive differences in tissue scattering. .

[0068] In the construction of the optimization function as shown in Equation (3), not only is the improvement of the overall sensitivity of the two bands considered, but also upper and lower limits of spectral distance are introduced to ensure that the selected wavelength combination is consistent with the physical assumptions of the subsequent inversion process. Specifically, when the two candidate wavelengths are too close, their ability to distinguish the oxygenation state may be insufficient, and the difference information between the two bands is not obvious, which is not conducive to the subsequent local optical path ratio estimation and inversion stability; while when the two candidate wavelengths are too far apart, the difference in the penetration depth and scattering characteristics of different wavelengths in the tissue may increase, thereby weakening the applicability of establishing local correction relationships based on the similar propagation conditions of the two bands. Through the above constraint solution process, wavelength pairs that take into account both sensitivity and physical adaptability can be selected from the candidate wavelength combinations to serve as the center wavelength configuration for subsequent two-band imaging.

[0069] The adaptive wavelength selection mechanism introduced in this application establishes a dynamic wavelength configuration strategy tailored to individual physiological differences. By incorporating the effects of melanin absorption, tissue thickness-related scattering, and hemoglobin extinction characteristics into the candidate wavelength evaluation process, and further combining spectral distance constraints to achieve dual-band optimization, the determined center wavelength not only possesses good initial detection sensitivity but also better adapts to the applicable conditions of subsequent local optical path correction and dual-band inversion processes. Therefore, the measurement adaptability and spatial imaging stability of the method in this application can be improved under different skin colors, different tissue thicknesses, and complex local tissue conditions.

[0070] Figure 2 A flowchart illustrating an example of obtaining a normalized dual-band image and its corresponding region of interest using a method according to an embodiment of this application is shown.

[0071] like Figure 2 As shown, in step S210, a pre-acquired dark field image and a whiteboard reflection image acquired based on a standard reference whiteboard are obtained.

[0072] In practical implementation, the dark-field image can be a baseline response image acquired by the imaging sensor under conditions where the excitation source is turned off and ambient light is blocked from entering the imaging field of view as much as possible. It is primarily used to characterize the dark current distribution and spatial characteristics of thermal noise of the imaging sensor itself. The whiteboard reflection image can be acquired using a standard reference whiteboard with known and high reflectivity, under the same lighting and imaging conditions as the actual measurement. It is primarily used to characterize the spatial illumination distribution of the light source within the two-dimensional field of view. In some embodiments, the standard reference whiteboard can be a reference plate made of a high-reflectivity diffuse reflection material, such as a polytetrafluoroethylene (PTFE) reference plate, but is not limited to this. Thus, the system can obtain two types of reference data reflecting the baseline noise level and spatial illumination distribution characteristics of the optical system before formal measurement. The dark-field image provides a sensor response benchmark under zero-input conditions, while the whiteboard reflection image provides a spatial response benchmark under ideal high-reflectivity conditions.

[0073] In step S220, for each image frame in the dual-band diffuse reflection original image sequence, flat field correction processing is performed using the dark field image and the whiteboard reflection image to eliminate non-uniform illumination and sensor dark current interference, and generate the corresponding corrected image sequence.

[0074] The flat-field correction process includes: subtracting the dark-field intensity of the dark-field image from the original reflectance intensity of the original image frame, and then dividing by the difference between the whiteboard reflectance intensity and the dark-field intensity of the whiteboard reflectance image to obtain the corresponding normalized reflectance intensity. For example, the normalized reflectance intensity can be calculated using the following formula:

[0075] Equation (4)

[0076] In the formula, Indicates in band Next moment In pixel coordinates Normalized reflection intensity at that location, For any one of two specific bands, This indicates the original image sequence of dual-band diffuse reflection in the band. Next moment In pixel coordinates The original reflection intensity at that location, Represents the dark field image in pixel coordinates Dark field intensity at that location This indicates the whiteboard reflection image in the band. Below pixel coordinates The intensity of whiteboard reflection at that location.

[0077] In equation (4), the flat-field correction process reduces the influence of sensor baseline noise by subtracting the dark field intensity in the numerator and normalizes the spatial illumination non-uniformity of the light source by introducing a white plate reference response in the denominator. After flat-field correction, the reflection intensity at different pixel positions can be compared under a unified benchmark, which helps to make the subsequent dual-band reflection information reflect more the optical response differences of the tissue under test itself, rather than the positional deviation caused by the hardware system.

[0078] In step S230, subpixel-level micro-motion compensation based on phase correlation between adjacent frames is performed on the corrected image sequence, and the cardiac prior information of the tissue region to be tested is initially estimated from the compensated corrected image sequence.

[0079] It should be noted that in actual non-contact measurement scenarios, the object being measured often exhibits small movements caused by breathing fluctuations, slight tremors, or posture adjustments. If these minute movements are not compensated for, the same pixel location in consecutive time frames may correspond to different tissue parts, thus affecting subsequent temporal pulsation signal extraction and spatial distribution analysis.

[0080] Therefore, this embodiment can estimate the global translation offset of consecutive image frames based on the phase correlation between adjacent frames, and perform sub-pixel level translation compensation for the current frame accordingly.

[0081] In some implementations, for adjacent consecutive frames... and The spatial offset can be solved by calculating the normalized cross power spectrum of the two in the frequency domain:

[0082] Equation (5)

[0083] In the formula, Represents the two-dimensional discrete Fourier transform. This represents the complex conjugate of the corresponding two-dimensional discrete Fourier transform. The coordinates are in the frequency domain. After performing an inverse Fourier transform on the cross power spectrum, the corresponding peak position can be obtained in the spatial domain. This peak position can characterize the global translational offset between adjacent frames. After interpolating and translating the image frames based on the offset, the spatial alignment between adjacent frames can be improved. After completing the micro-motion compensation, the system can also perform spectral analysis on the pixel intensity time series of the entire field of view or a specific tissue region to obtain the dominant frequency peak reflecting the physiological pulsation frequency of the current period, and use it as the a priori information for subsequent filtering processing. Thus, the impact of micro-motion on the stability of the time series signal can be reduced, and a priori basis can be provided for the subsequent extraction of pulsation-related frequency bands.

[0084] In step S240, an adaptive bandpass filter that tracks the a priori information of the heart is used to extract the frequency band of the tissue pulsation signal to filter out ambient light noise and low-frequency motion artifacts, thereby generating a spatiotemporally aligned normalized dual-band image.

[0085] Since background light variations in actual imaging environments are typically concentrated in lower frequency bands or near specific power frequencies, while residual motion artifacts may be distributed across a wider frequency band, using a filtering method with fixed passband boundaries may result in the weakening of useful pulsation components or the retention of significant amounts of irrelevant noise. Therefore, this embodiment utilizes the cardiac a priori information obtained in step S230 to adaptively adjust the center frequency and passband range of the bandpass filter, making the filter's passband closer to the actual physiological pulsation range of the subject being measured.

[0086] In some implementations, the filter passband can be set within a preset tolerance range near the heart rate fundamental frequency. For example, a passband boundary can be set with a certain percentage of fluctuation above and below the current fundamental frequency, and pixel-by-pixel temporal filtering can be performed on the compensated and corrected image sequence accordingly. After this processing, the effective pulsation components related to local microcirculation perfusion changes in the image sequence can be preserved, while the influence of ambient light disturbances and low-frequency motion trend terms can be further suppressed. Through this step, a normalized dual-band image with high temporal consistency and spatial alignment can be generated, thereby providing more stable input data for subsequent local optical path estimation and blood oxygenation inversion.

[0087] In step S250, the channel reflectance ratio of the normalized dual-band image under two specific bands is extracted as the skin color feature ratio, and combined with the edge detection operator to determine the initial contour of the epidermal tissue. Within the initial contour, a superpixel segmentation algorithm is used to perform structural clustering of pixels with homogeneous optical scattering characteristics to extract the set of pixels containing effective physiological tissue as the region of interest, thereby obtaining the normalized dual-band image and its corresponding region of interest.

[0088] Here, clothing, dressings, background areas, and other irrelevant regions are removed from the complex imaging field of view, and the effective physiological tissue regions that truly participate in subsequent optical analysis are extracted. Specifically, since physiological tissues typically exhibit specific diffuse reflectance ratio distribution characteristics under dual-band conditions, the channel reflectance ratios under two specific bands can be used as a preliminary basis for distinguishing physiological tissues from non-physiological backgrounds. Simultaneously, combined with edge detection results, the initial geometric contour range of epidermal tissue in the image can be further determined.

[0089] After defining the initial contour, considering that different locations on the tissue surface may still have differences in local scattering characteristics and texture distribution, this embodiment further introduces a superpixel segmentation algorithm. Within the initial contour range, this algorithm comprehensively considers the spatial proximity relationship between pixels and the similarity of dual-band optical features, aggregating pixels into several structural units with strong homogeneous optical scattering characteristics. Subsequently, a set of pixels that meet the preset physiological tissue characteristic conditions can be selected from these structural units and identified as the region of interest. This not only eliminates obviously irrelevant backgrounds but also preserves relatively clear local texture boundaries and scattering structure partitions within the tissue.

[0090] Through the embodiments of this application, a preprocessing and region of interest extraction mechanism for dual-band diffuse reflection imaging is established by integrating flat field correction, micro-motion compensation, temporal filtering, and structured region segmentation. This mechanism can suppress hardware response deviations, environmental noise, and motion disturbances in the original image sequence at a uniform scale, and further extract target tissue regions with high physiological relevance. This improves the spatial comparability and temporal stability of dual-band reflection signals, providing higher quality input data for subsequent local optical path correction and blood oxygen distribution inversion.

[0091] Regarding the implementation details of constructing the local optical path correction matrix covering the region of interest in step S130, in some examples of embodiments of this application, it can be specifically achieved through the following optimization steps that combine diffuse reflection theory and image spatial analysis.

[0092] First, for each target pixel in the region of interest, the normalized reflection intensity of each neighboring pixel in the local neighborhood of the target pixel is extracted in two specific bands, and local logarithmic transformation is performed on each to obtain the local logarithmic reflection intensity distribution corresponding to the two specific bands.

[0093] In practice, considering that photons are subjected to both absorption and scattering during propagation within biological tissues, their reflection intensity typically exhibits a nonlinear attenuation characteristic related to the propagation distance on a macroscopic scale. Directly using the original reflection intensity for local spatial analysis would be detrimental to extracting linear features corresponding to the local attenuation trend. Therefore, in this step, a natural logarithmic transformation is performed on the normalized reflection intensity. For example, the local logarithmic reflection intensity can be obtained using the following formula:

[0094] Equation (6)

[0095] In the formula, Represents the coordinates of neighboring pixels In the band Local logarithmic reflection intensity; This represents the normalized reflection intensity at the corresponding position in the preprocessing output. The above transformation converts the exponential attenuation relationship in the original reflection intensity into a numerical distribution more suitable for local linear modeling, while compressing the dynamic range of the light intensity signal. This facilitates the subsequent extraction of image gradient information characterizing local optical attenuation properties.

[0096] Then, for each specific band in the two specific bands, a spatial linear regression model is established with the Euclidean distance from each neighboring pixel in the local neighborhood to the target pixel as the independent variable and the local logarithmic reflection intensity of the corresponding neighboring pixel as the dependent variable. The slope obtained by fitting is determined as the reflection intensity attenuation slope corresponding to the specific band. The reflection intensity attenuation slope is used to characterize the image gradient information.

[0097] In a microscopic local field of view, the following spatial linear regression model can be used to fit the trend of logarithmic reflection intensity as a function of distance:

[0098] Equation (7)

[0099] In the formula, Represents neighboring pixels To target pixel The Euclidean distance; Let be the slope of the reflection intensity attenuation to be fitted; This is the fitting constant term. Solving the model using the least squares method or other regression algorithms yields the attenuation slope corresponding to each specific band. .

[0100] In equation (7), the attenuation slope can characterize the attenuation trend of photon energy in a local region as the lateral diffusion distance increases, which corresponds to the local image gradient information. Compared with simple two-point difference calculation, using a local neighborhood window for spatial regression fitting can reduce the influence of single-pixel random noise and local small texture abrupt changes on the results to a certain extent, thereby obtaining more stable local attenuation features.

[0101] Subsequently, based on the diffuse reflection approximation theory, the initial optical path ratio estimate of the target pixel is determined according to the ratio of the reflection intensity attenuation slopes corresponding to the two specific wavebands.

[0102] According to the diffuse reflection approximation theory, there is a corresponding relationship between the attenuation slope of the local logarithmic reflection intensity and the effective attenuation coefficient of the tissue. Considering that the two wavelengths in dual-band imaging are usually quite close, their scattering components within the same local homogeneous tissue region have a strong correlation. Therefore, this embodiment weakens the influence of unknown local scattering by constructing a ratio relationship of the dual-band attenuation slopes. For example, the initial optical path ratio estimate can be determined by the following formula:

[0103] Equation (8)

[0104] In the formula, Indicates target pixel The initial optical path ratio estimate at the location; and These represent the attenuation slopes of the reflection intensity corresponding to the first and second specific wavebands, respectively. This ratio can mitigate the direct influence of the local absolute scattering coefficient and surface reflection amplitude on the results to some extent, thereby extracting proportional information that more closely approximates the local light propagation differences between the two wavebands, providing initial results for the subsequent stable estimation of the effective optical path ratio.

[0105] Subsequently, the initial optical path ratio estimate of each pixel in the region of interest is subjected to iterative regularization smoothing. The iterative regularization smoothing includes: determining Gaussian spatial distance weights based on the spatial distance between pixels in the local neighborhood, determining similarity weights based on image edge features, and using the Gaussian spatial distance weights and similarity weights to perform joint weighted filtering on the initial optical path ratio estimate.

[0106] Considering that although the initial optical path ratio estimate obtained in the preceding steps already reflects local light propagation differences, discrete fluctuations or local discontinuities may still occur at blood vessel boundaries, in areas of abrupt tissue structure changes, or in areas with strong local noise, this embodiment further introduces a jointly weighted iterative regularization smoothing mechanism to perform edge-preserving optimization on the initial optical path ratio estimate. For example, the iterative update process can be expressed as:

[0107] Equation (9)

[0108] Equation (10)

[0109] In the formula, and They represent the first Second and third Optical path ratio of the next iteration; For filtering neighborhood; For joint weights; and These are the standard deviation parameters that control the smoothness of spatial distance and the smoothness of edge similarity, respectively. It represents the difference in edge features between neighboring pixels and the target pixel in a dual-band image, such as gradient magnitude difference or intensity difference.

[0110] By combining the weighting methods of Equations (9) and (10) above, the spatial distance weight is beneficial to improving the continuity of optical path ratio estimation in similar regions, while the similarity weight is beneficial to suppressing excessive smoothing when crossing blood vessel boundaries or tissue structure boundaries, thereby achieving a balance between denoising and edge preservation.

[0111] Furthermore, the optical path ratio value after iterative convergence is determined as the effective optical path ratio of the corresponding pixel, and a local optical path correction matrix is ​​constructed from the effective optical path ratio of each pixel in the region of interest.

[0112] In some implementations, when the difference in optical path ratio updates between two consecutive iterations is lower than a preset threshold, the iteration process is considered to have converged, and the iteration is stopped. The converged result is then determined as the final effective optical path ratio. Furthermore, the effective optical path ratio of each pixel within the region of interest is organized according to its spatial location to form a two-dimensional local optical path correction matrix. This matrix corresponds one-to-one with the spatial location of the original image and can serve as the basis for point-by-point optical path compensation in the subsequent dual-band inversion model, thereby improving the adaptability of the subsequent blood oxygenation inversion process to local tissue scattering differences, thickness variations, and microstructural inhomogeneities.

[0113] This embodiment employs a technical chain of "local logarithmic transformation—spatial linear regression—dual-band slope ratio estimation—edge-preserving iterative smoothing" to construct a local optical path adaptive correction mechanism. This mechanism utilizes local spatial variation information from a dual-band normalized image to estimate the dual-band propagation differences at each pixel location within the region of interest, and further forms a local optical path correction matrix corresponding to the image's spatial location. This provides a compensation basis that better conforms to the local tissue optical characteristics for subsequent pixel-level blood oxygenation inversion, thereby improving the stability and spatial consistency of two-dimensional blood oxygenation imaging results.

[0114] Figure 3 A flowchart illustrating an example of calculating the preliminary blood oxygen saturation of a pixel using a dual-band inversion model according to an embodiment of this application is shown.

[0115] like Figure 3 As shown, in step S310, for each pixel in the region of interest, the normalized reflectance intensity under the first and second specific bands in two specific bands is extracted, and the corresponding tissue absorbance is calculated based on the negative logarithmic transformation of the normalized reflectance intensity.

[0116] In practical implementation, based on the transmission law of light in an absorbing medium, when photons penetrate or enter biological tissue containing absorbing components, the intensity of their emitted or reflected light typically exhibits an exponential decay characteristic on a macroscopic scale. If linear reflection intensity is directly used for subsequent inversion, it is not conducive to establishing a clear correspondence between tissue absorption characteristics and internal chromophore concentration. Therefore, in this embodiment, a negative natural logarithmic transformation is performed on the normalized reflection intensity to map the reflection intensity to tissue absorbance. For example, pixels in any specific wavelength band... Tissue absorbance It can be calculated using the following formula:

[0117] Equation (11)

[0118] In the formula, Represents pixel coordinates In the band Normalized reflectance intensity; This represents the absorbance of the corresponding tissue after negative logarithmic transformation.

[0119] In step S320, the ratio of tissue absorbance corresponding to the first specific band to tissue absorbance corresponding to the second specific band is determined as the reflection attenuation ratio of the corresponding pixel.

[0120] In non-contact imaging scenarios, besides blood absorption, factors such as tissue surface roughness, incident geometry, and localized high-brightness reflection can also collectively affect absorbance. Considering that these non-specific factors are typically strongly correlated in two closely spaced specific wavelength bands, this embodiment reduces the impact of such common perturbations by calculating the ratio of tissue absorbance in both wavelength bands. For example, the reflection attenuation ratio of the corresponding pixel... It can be represented as:

[0121] Equation (12)

[0122] By using this ratio calculation, the influence of wavelength-independent absolute light intensity fluctuations and system baseline offset on subsequent solution results can be reduced to a certain extent, making the obtained reflection attenuation ratio more representative of the relative absorption differences within the tissue under dual-band conditions.

[0123] In step S330, the reflection attenuation ratio of the corresponding pixel and the effective optical path ratio of the pixel in the local optical path correction matrix are substituted into the dual-band inversion model.

[0124] Here, the dual-band inversion model is configured to establish an inversion function mapping relationship between the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the first and second specific bands and the preliminary blood oxygen saturation, based on the improved Lambert-Beer law. The preliminary blood oxygen saturation of the corresponding pixel is solved by eliminating the total hemoglobin concentration term.

[0125] More specifically, the inversion function mapping relationship is configured as follows: the initial blood oxygen saturation is equal to the ratio of the target numerator to the target denominator; the target numerator is obtained by subtracting the extinction coefficient of deoxyhemoglobin in the first specific band from the product of the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficient of deoxyhemoglobin in the second specific band; the target denominator is obtained by subtracting the product of the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the second specific band from the product of the reflection attenuation ratio, the effective optical path ratio, and the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the second specific band; and the effective optical path ratio is the ratio of the effective optical path corresponding to the second specific band to the effective optical path corresponding to the first specific band.

[0126] In some implementations, based on a modified Lambert-Beer law, any band Tissue absorbance It can be determined by the total hemoglobin concentration Blood oxygen saturation and the effective optical path at that wavelength The union is represented as:

[0127] Equation (13)

[0128] In the formula, This represents the total hemoglobin concentration at the corresponding pixel location, which is an unknown quantity in this inversion process. Substituting the above relationships into the first specific band... Second specific band In the corresponding absorbance expression, and combined with the reflection attenuation ratio obtained in step S320, Since the reflection attenuation ratio is defined as the ratio of absorbance in two wavelength bands, the total hemoglobin concentration term is included in both the numerator and denominator. It was removed.

[0129] After algebraic rearrangement and merging of like terms, the inversion function mapping is ultimately configured as follows: the initial blood oxygen saturation equals the ratio of the target numerator to the target denominator. For example, the initial blood oxygen saturation of the corresponding pixel... It can be obtained by calculation using the following formula:

[0130] Equation (14)

[0131] In the formula, Represents pixel coordinates Preliminary blood oxygen saturation at the site; Represents pixel coordinates The reflection attenuation ratio at that location; This represents the effective optical path ratio corresponding to the local optical path correction matrix, and this effective optical path ratio... It is the ratio of the effective optical path length corresponding to the second specific band to the effective optical path length corresponding to the first specific band; and These represent oxyhemoglobin and deoxyhemoglobin respectively in the first specific wavelength band. The extinction coefficient at the specified value; and They respectively indicate that they are in the second specific band. The extinction coefficient below.

[0132] Combining equation (14), it can be seen that the target numerator term specifically corresponds to the numerator part, that is, the reflection attenuation ratio. Effective optical path ratio And the extinction coefficient of deoxyhemoglobin in the second specific wavelength band The product of the products, minus the extinction coefficient of deoxyhemoglobin in the first specific wavelength band. get.

[0133] The target denominator term specifically corresponds to the denominator part, which is the difference in extinction coefficients between oxyhemoglobin and deoxyhemoglobin in the first specific wavelength band. Subtract the reflection attenuation ratio Effective optical path ratio And the difference in extinction coefficients between oxyhemoglobin and deoxyhemoglobin in the second specific wavelength band. The product of the products is obtained.

[0134] As can be seen from the above inversion relationship, the initial blood oxygen saturation is jointly determined by the reflection attenuation ratio, the local effective optical path ratio, and the extinction characteristics of the two hemoglobins in the dual-band. The reflection attenuation ratio characterizes the local dual-band absorption difference, the effective optical path ratio reflects the difference in the propagation path of different wavelengths of light in the local tissue, and the dual-band extinction coefficient provides the a priori absorption characteristics of hemoglobin in each band. This embodiment integrates these factors into the same inversion model through a physical inversion link, establishing a direct functional mapping between tissue surface optical observations and underlying blood oxygenation status at the pixel level. Compared to processing methods that rely solely on fixed empirical calibration relationships, this embodiment introduces local optical path correction information during the inversion process, which can, to some extent, reduce the impact of non-specific optical perturbations on the tissue surface and the unknown total hemoglobin concentration on blood oxygen inversion. Therefore, it can provide initial blood oxygen results with pixel-level quantitative significance for subsequent two-dimensional blood oxygen distribution optimization, thereby improving the accuracy and spatial resolution of blood oxygenation imaging under complex tissue conditions.

[0135] The dual-band inversion model proposed in this application establishes a physical inversion chain for calculating preliminary blood oxygen saturation. This chain involves normalized reflectance intensity to tissue absorbance, absorbance ratio to reflectance attenuation ratio, and then combining effective optical path ratio and hemoglobin extinction coefficient. This approach can mitigate the impact of non-specific optical perturbations on tissue surface and unknown total hemoglobin concentration on blood oxygen saturation calculations. Furthermore, by combining the local optical path correction results obtained in previous steps, it compensates for differences in dual-band propagation at different locations. Therefore, it provides initial blood oxygen results with pixel-level quantitative significance for subsequent two-dimensional blood oxygen distribution optimization, thereby improving the accuracy and spatial resolution of blood oxygen imaging under complex tissue conditions.

[0136] Regarding the implementation details of generating the two-dimensional blood oxygen saturation distribution map corresponding to the region of interest through spatial regularization optimization in step S150, in some examples of the embodiments of this application, it can be specifically achieved by constructing a global variational framework and adopting an iterative optimization method.

[0137] First, a global variational objective function is constructed, comprising a data fidelity term and a spatial smoothing regularization term with edge-preserving weights. The data fidelity term is used to characterize the consistency difference between the measured value of the reflection attenuation ratio of the corresponding pixel and the theoretical reflection attenuation ratio derived from the dual-band inversion model based on the current blood oxygen saturation and effective optical path ratio through the sum of squared errors. The spatial smoothing regularization term is used to apply a weighted penalty to the square of the blood oxygen saturation difference between adjacent pixels under the joint adjustment of the local texture edge-preserving weight and the global regularization weight coefficient. The edge-preserving weight is dynamically determined based on the local texture gradient of the normalized dual-band image and is used to reduce the over-smoothing of the tissue boundary region.

[0138] For example, the global variational objective function It can be expressed by the following formula:

[0139] Equation (15)

[0140] In the formula, pixel coordinates The reflection attenuation ratio at that location is used as the actual observed data item in the objective function; To determine based on current blood oxygen saturation And effective optical path ratio The theoretical reflection attenuation ratio determined by the dual-band inversion model; These are global regularization weights used to control the overall smoothness. To include all adjacent pixel pairs and edge set; and These represent adjacent pixel pairs in coordinates. and Blood oxygen saturation at the site; Edge-preserving weights are dynamically determined based on the local texture gradient of a normalized dual-band image, used to reduce over-smoothing of tissue boundary regions.

[0141] In some implementations, edge-preserving weights Further quantification can be performed using the following attenuation function based on local intensity differences:

[0142] Equation (16)

[0143] In the formula, and These represent the joint feature vectors of two adjacent pixels in a normalized dual-band image, composed of the intensities of two specific band channels. The L2 norm of a vector. To control the smoothing attenuation constant for edge sensitivity, the above settings result in a relatively large edge preservation weight when the intensity difference between adjacent pixels is small, which helps to enhance the smoothing constraint within homogeneous tissue regions. Conversely, when the intensity difference between adjacent pixels is large, the edge preservation weight is relatively reduced, which helps to reduce over-smoothing of tissue boundaries, blood vessel boundaries, or lesion boundaries. Therefore, image structure priors can be combined with the blood oxygen distribution optimization process, allowing the results to better preserve local boundary features while suppressing noise.

[0144] Then, using the initial blood oxygen saturation corresponding to each pixel as the initial iteration value, the gradient descent algorithm is used to minimize the global variational objective function and update iteratively. The blood oxygen saturation of the next iteration is obtained by updating the blood oxygen saturation of the current iteration along the negative gradient direction of the global variational objective function at the current blood oxygen saturation, combined with the preset iteration step size.

[0145] For example, minimizing the iterative update can be expressed as follows:

[0146] Equation (17)

[0147] In the formula, and They represent the first Second and third Blood oxygen saturation in the next iteration The iteration step size, This represents the partial derivative gradient of the global variational objective function at the blood oxygen saturation level in the current iteration.

[0148] In practice, the preliminary blood oxygen saturation obtained by independent pixel-by-pixel inversion in the preceding steps can be used as... Initialization is performed. Since this initial value already includes the dual-band inversion model and local optical path correction information, it typically provides a reasonable starting state for subsequent variational optimization. Subsequently, in each iteration, the gradient descent algorithm updates the blood oxygen value of each pixel according to the direction indicated by the current partial derivative gradient, thus gradually seeking a balance between "meeting the constraints of actual observation data" and "maintaining the smoothness of spatial distribution." Iteration step size. This is used to control the magnitude of each update, balancing the convergence speed and numerical stability of the optimization process. Through this iterative update process, random noise and local discrete disturbances in the initial blood oxygen distribution can be gradually reduced, while maintaining consistency with the underlying physical inversion model.

[0149] Furthermore, when the decrease in loss of the global variational objective function is lower than the preset convergence threshold or the maximum number of iterations is reached, the iteration is stopped, and the final converged blood oxygen saturation result is determined as the two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

[0150] In some implementations, convergence criteria based on the rate of change of loss and a maximum iteration constraint can be set simultaneously to ensure good controllability of the algorithm under different input quality conditions. Once the optimization process meets the stopping condition, the output blood oxygen saturation result has achieved a balance between physical observation consistency and spatial structural continuity, and can serve as a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest. Compared to the preliminary blood oxygen results obtained by solving pixel-by-pixel independently, the two-dimensional blood oxygen distribution map optimized by spatial regularization typically better suppresses local isolated noise and random artifacts, while also better preserving the spatial representation of tissue boundaries and lesion boundaries, thus providing a more stable basis for subsequent imaging observation and auxiliary analysis.

[0151] Through the embodiments of this application, pixel-level preliminary blood oxygenation results are incorporated into a global variational optimization framework. Based on data fidelity constraints, an edge-preserving spatial smoothing mechanism is introduced, thereby establishing a two-dimensional blood oxygenation distribution optimization method that takes into account both physical consistency and spatial structural continuity. This method can reduce the loss of local boundary information while suppressing noise disturbances, which is beneficial to improving the stability, interpretability, and spatial representation ability of the final two-dimensional blood oxygen saturation distribution map.

[0152] In a further preferred embodiment of this application, a variable weight regularization mechanism based on hardware resolution adaptation and an optical path-oxygenation coupling update mechanism can be introduced for the minimization iterative update process in step S150. This optimization scheme can improve the preservation of microvascular structures and local oxygenation mutation regions while maintaining overall denoising capabilities, and better coordinate physical model parameters with physiological parameters during iteration. Specific implementation details and technical logic may include the following.

[0153] Specifically, first, the half-width and height of the point spread function corresponding to the current optical system are determined, and then the normalized dual-band image is processed using a multi-scale vascular enhancement filtering algorithm to determine the local vascular structure scale corresponding to each pixel in the region of interest.

[0154] In practical implementation, the half-width at half-maximum (FWHM) of the point spread function (PSF) can be used as a hardware parameter characterizing the spatial resolution capability of the imaging system. This parameter can be obtained in advance through system calibration and reflects the current optical system's imaging resolution limit for microstructures. Based on this, this embodiment can utilize multi-scale vessel enhancement filtering algorithms, such as Frangi filtering, multi-scale analysis methods based on the Hessian matrix, or other algorithms capable of characterizing the response of slender structures, to process the normalized dual-band image. By analyzing the image structural response in multiple scale spaces, the local vascular structure scale or local tissue structure feature size corresponding to each pixel position within the region of interest can be obtained.

[0155] Therefore, a correspondence can be established between the hardware resolution capability of the imaging system and the scale of the local anatomical structure of the target tissue. In other words, the system can identify which regions have a structural scale much larger than the resolution limit of the imaging system, and which regions are close to or below that resolution limit.

[0156] Then, based on the scale ratio between the local vascular structure scale and the half-width at half-maximum (WHM) of the point spread function, the pixel-wise spatial weighting regularization coefficients are determined. Furthermore, when calculating the spatial smoothing regularization term and its partial derivative gradient in the global variational objective function, pixel-wise spatial weighted regularization coefficients are used. Replace the regularization weight coefficients; where, for pixels whose local vascular structure scale is smaller than the half-width and half-height of the point spread function, the pixel-wise spatial weighted regularization coefficient of the corresponding pixel is reduced according to the scale ratio through a monotonically decreasing mapping function. .

[0157] In this process, to avoid over-smoothing of fine structures due to traditional globally uniform smoothing weights, this embodiment introduces pixel-wise variable weighting regularization coefficient mapping logic. For example, pixel-wise spatial variable weighting regularization coefficients... It can be determined by the following monotonically decreasing mapping function:

[0158] Equation (18)

[0159] In the formula, As the base weights for global regularization, pixel coordinates Local vascular structure scale. Let the half-width and height of the point spread function be the area of ​​the point spread function. This is the steepness adjustment index.

[0160] As can be seen from the above relationships, when the scale of local vascular structures is significantly larger than the system's resolving power, It can maintain a high value, thus maintaining a strong smoothing constraint in the region, thereby improving the denoising effect in the background region or homogeneous region; while when the scale of the local vascular structure is smaller than or close to the half-width at half-maximum of the point spread function, This will decrease accordingly, thereby reducing the smoothing intensity at that location. Through this pixel-by-pixel adjustment mechanism, spatial regularization can maintain good noise suppression capabilities in the macro background region, while avoiding excessive smoothing of microvessels or local oxygenation mutation regions during the optimization process. This is beneficial for improving the spatial preservation ability of the final two-dimensional blood oxygen distribution results for microstructures.

[0161] Subsequently, a coupled update mechanism between local optical path length and tissue oxygenation status is established during the iterative update process: in the current iteration update, based on the current blood oxygen saturation... And the reflection attenuation ratio of the corresponding pixel, combined with the dual-band inversion model to determine the current effective optical path ratio. Perform synchronous correction to obtain the effective optical path ratio after the current round of updates. .

[0162] It should be noted that in the previous implementation, the effective optical path ratio, as an important input parameter for dual-band inversion, is usually estimated by the local image gradient and diffuse reflection diffusion model; however, in a further preferred implementation, this embodiment considers that the tissue absorption state and local propagation conditions may interact during the iteration process, therefore, the effective optical path ratio is also synchronously corrected while the blood oxygen distribution is continuously updated. Specifically, in the first... Blood oxygen distribution was obtained through rounds of iteration. Then, the effective optical path ratio can be updated to match the current blood oxygenation state by combining the current pixel's reflection attenuation ratio and the dual-band inversion model. For example, the corrected effective optical path ratio... The aforementioned inversion model can be rearranged as follows:

[0163] Equation (19)

[0164] This synchronous correction step ensures that the effective optical path ratio is no longer a fixed parameter after the initial estimate, but is dynamically updated during the iteration process in conjunction with the current oxygenation state. This helps to gradually align the local light propagation conditions with the current blood oxygenation state during the solution process, thereby reducing the continued impact of the initial optical path ratio estimation bias on subsequent optimization results.

[0165] Furthermore, when calculating the global variational objective function and its partial derivative gradient for the next iteration, the effective optical path ratio updated in the current iteration is used. Replace the effective optical path ratio in the global variational objective function .

[0166] Here, the revised version will be... Resubmit the data fidelity item for use in the first... The gradient descent update calculation is performed in each round. In this way, the direction of the blood oxygen distribution update in the next round is no longer determined solely by the fixed optical path parameters of the previous round, but is based on the latest optical path correction results. Through this closed-loop feedback process, the global variational objective function can simultaneously consider the current blood oxygen state and the current local propagation conditions in each iteration, thereby achieving better coordination between data consistency, spatial structure constraints, and local optical propagation consistency in the optimization process.

[0167] In some implementations, the combined application of the aforementioned pixel-by-pixel spatial weighting regularization mechanism and effective optical path ratio synchronous correction mechanism can maintain good noise suppression capabilities in the macroscopic background region during the two-dimensional blood oxygenation distribution optimization process, while maintaining good structural resolution capabilities in microvascular regions, lesion boundary regions, or local blood supply abrupt change regions. Therefore, the final two-dimensional blood oxygenation distribution map can be further improved in terms of spatial continuity, preservation of local boundaries, and consistency of physical interpretation.

[0168] This embodiment introduces a hardware resolution-aware spatial weighting regularization mechanism and an optical path-oxygenation coupling update mechanism. On the one hand, it adaptively adjusts the spatial smoothing intensity based on the relationship between local structural scale and system resolution to balance background denoising and preservation of fine structures. On the other hand, by synchronously updating the effective optical path ratio during the iteration process, a closed-loop correction relationship is formed between blood oxygen inversion and optical path compensation. Therefore, the stability, boundary fidelity, and physical consistency of the final two-dimensional blood oxygen saturation distribution map in complex tissue scenarios can be further improved.

[0169] Figure 4 A schematic diagram illustrating the system operation mechanism of an example of a blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown.

[0170] like Figure 4 As shown, the imaging system of this embodiment is equipped with a high-speed diffuse reflection camera and a dual-band uniform light source at the hardware acquisition end, for acquiring the original dual-band diffuse reflection image sequence of the tissue region under test in a specific band. During this acquisition stage, the system uses the adaptive wavelength selection module in the scheduling and constraint framework to dynamically schedule the dual-band uniform light source in combination with the individual physiological characteristic parameters of the tissue under test, thereby determining a suitable combination of center wavelengths and controlling the acquisition of the original sequence.

[0171] The acquired raw image sequences then enter the data processing chain. First, image preprocessing and normalization modules reduce interference from system noise and non-uniform illumination. Then, a multi-scale ROI (Region of Interest) segmentation module is used to remove non-target background, extracting the ROI containing valid physiological tissue and its corresponding normalized image. Next, LPDE (Local Pathlength Differential Estimation) is performed on the ROI to estimate local optical propagation differences based on the local spatial variation features of the image, and the corresponding correction matrix is ​​output. This correction matrix, along with the normalized image, is further input into a dual-band pixel-level inversion model. Based on algebraic mapping relationships, the effects of local optical absorption and scattering are decoupled, and the pixel-level values ​​at each pixel location are initially calculated and output. (Oxygen Saturation).

[0172] Considering the potential spatial discrepancies and random disturbances that pixel-level independent inversion may introduce, the system further introduces a top-level spatial regularization constraint module to perform smoothing optimization on the initial pixel-level blood oxygenation results. As shown by the dashed feedback link from the spatial regularization constraint module back to the dual-band pixel-level inversion model in the figure, this smoothing optimization process establishes a coupled update mechanism between local optical path length and tissue oxygenation state during iterative updates. By combining the dual-band inversion model to synchronously correct the effective optical path ratio, the local light propagation conditions and blood oxygenation state gradually converge during the solution process. After spatial regularization smoothing optimization, the system outputs a two-dimensional blood oxygen saturation heatmap that balances physical observation consistency and spatial structural continuity, and can simultaneously generate statistical indicators and analysis reports, thereby realizing two-dimensional spatial characterization and quantitative analysis of the blood oxygenation state of the target tissue region.

[0173] To fully verify the effectiveness and applicability of the blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping provided in this application embodiment under complex physiological conditions and non-ideal environments, this embodiment constructs a highly realistic biological tissue optical simulation model and conducts systematic comparative experiments and analyses from three dimensions: spatial resolution, robustness to individual physical characteristics, and algorithm performance boundaries.

[0174] Specifically, in terms of experimental setup and biomimetic model construction, unlike traditional simple mathematical models that assume uniform optical distribution, this embodiment uses the Monte Carlo Modeling of Light Transport in Multi-Layered Tissues (MCML) algorithm to construct a three-dimensional multilayered skin tissue biomimetic phantom. This model, from top to bottom, includes the epidermis, dermis, and subcutaneous tissue layer. Different concentrations of melanin are dynamically configured in the epidermis, and microvascular networks with different diameters and topological orientations are embedded in the dermis to accurately simulate the non-uniform optical scattering and absorption characteristics of real biological tissues.

[0175] In a specific simulation scenario, this embodiment simulates a localized ischemic lesion with physiological gradients (such as an early ulcer or pressure sore). The baseline oxygen saturation of the background normal dermal tissue is set at 96%, while the central ischemic area exhibits a spatial distribution that gradually decreases to 85% from the periphery to the center. Furthermore, to maximize the replication of the realistic non-contact optical imaging physical process, this embodiment also simultaneously injects substrate speckle noise and detector Poisson noise at the photon collection end. This highly realistic non-ideal environment setting provides a clinically relevant test data foundation for subsequent verification of the noise resistance and edge preservation capabilities of the local optical path correction matrix and spatial regularization optimization mechanism in this application.

[0176] Figure 5 A comparative simulation diagram of an example of imaging in complex microvascular networks and localized ischemic areas is shown. The simulation diagram includes a high-resolution pseudo-color image comparison section at the top and a superimposed cross-sectional curve analysis section at the bottom, used to compare the imaging fidelity of the traditional global ratio method assuming a global constant optical path ratio with the method of the embodiments of this application in reconstructing the spatial distribution of blood oxygenation.

[0177] like Figure 5 The high-resolution pseudo-color image comparison section shows, in turn, the true value, the traditional global ratio method, and the method of the embodiments of this application. Two-dimensional spatial distribution. Due to the lack of effective compensation for local scattering differences, the traditional global ratio method exhibits significant halo artifacts at the microvascular edges and the boundary of ischemic areas, resulting in blurred physiological boundaries and poor global noise resistance. In contrast, the method in this application introduces an LPDE module to dynamically correct optical path distortion caused by non-uniform tissue scattering, thereby effectively suppressing halo artifacts and making the reconstructed blood oxygen distribution map highly consistent with the true value in visual morphology.

[0178] Furthermore, based on the single-row pixel values ​​corresponding to the extracted cross-section positions marked in the pseudo-color image above, Figure 5 Below, the superimposed cross-sectional curves of the three are plotted. The comparison results of the cross-sectional curves clearly show that in the tissue edge region where blood oxygenation changes abruptly, the blood oxygenation estimate of the traditional global ratio method exhibits significant overshoot and hysteresis, with its local error peak reaching ±4.5%. In contrast, the estimation curve of the method in this embodiment closely follows the spatial gradient curve of the true value, resulting in an overall RMSE (Root Mean Square Error) reduction of approximately 68%. This simulation comparison objectively verifies that the method in this embodiment, after synergistic local optical path correction and spatial regularization optimization, can smooth random noise while faithfully reproducing the physical distribution trend of microvessels and the true spatial structure of ischemic lesions.

[0179] Figure 6A schematic diagram illustrating the comparative experimental simulation results under individual differences in skin color is shown. This simulation diagram, presented in the form of a violin plot, compares the distribution of blood oxygen measurement errors between the traditional fixed-band method and the method of this application embodiment under interference from different individual physiological characteristics (i.e., different skin colors or melanin concentrations). Specifically, in the experimental setup, the epidermal melanin volume fraction was equidistantly adjusted between 1% and 15% in the simulated phantom, strictly corresponding to the objective Fitzpatrick skin color classification types I to VI, to systematically evaluate the applicability of the non-contact optical imaging device to different skin color populations and the robustness of the algorithm.

[0180] like Figure 6 The error distribution data shows that as melanin concentration increases, the absorption of diffuse reflected light signals by the tissue increases significantly, leading to a sharp attenuation of the effective optical signal. Traditional fixed-band methods (such as conventional RGB band or 610nm single-band measurements) cannot dynamically avoid interference from strong absorption bands due to the use of a uniform static light source configuration. This causes the median measurement error to increase sharply from about 2.1% in type I to about 7.8% in type VI, and the variance of the data distribution increases sharply (manifested as a significant elongation and divergence of the longitudinal distribution contour of the blue violin plot in the figure), resulting in a significant decrease in measurement accuracy under the dark skin model. In contrast, the adaptive wavelength selection mechanism adopted in this application embodiment, by combining the individual's melanin index and tissue thickness parameters to construct an optimization function before measurement, can dynamically evaluate the system noise and sensitivity factor under different candidate wavelengths, thereby accurately avoiding strong absorption interference and adaptively switching to a specific band combination with optimal penetration and photoelectric complementarity.

[0181] Therefore, as Figure 6 As shown in the red distribution outline corresponding to the method in the embodiments of this application, the median measurement error remained consistently below 2.5% across all Fitzpatrick skin color grades, and the variance of the error distribution was extremely small (manifested as the vertical distribution outline of the red violin diagram remained consistently compact). This experimental result objectively verifies that the adaptive wavelength selection strategy introduced in the embodiments of this application effectively overcomes the significant fluctuations in measurement accuracy of traditional non-contact blood oxygenation detection technology when dealing with people of different skin colors, demonstrating a strong adaptability and robustness to individual physiological differences.

[0182] Figure 7 This diagram illustrates a comparative simulation of an example in the study of algorithmic trade-offs and the limits of microscopic spatial resolution.

[0183] like Figure 7As shown above, the scatter plot of the Pareto Front illustrates the performance trade-off between spatial fidelity and computation time (in milliseconds). The plot marks data points from the traditional algorithm, the proposed solution, and non-Pareto optimal data points. It is clearly evident that while the variational space regularization module introduced in this embodiment significantly increases spatial fidelity, it also objectively increases computational overhead, as indicated by the arrows in the plot, showing the evolution of increased computation time versus increased fidelity. Furthermore, the points from this solution are closer to the Pareto Front. This objectively reveals the physical and performance boundaries between high spatial fidelity and extremely low computational latency when processing high-resolution sequences, providing a basis for trade-offs in system deployment and parameter adjustment under different terminal computing power conditions.

[0184] like Figure 7 As shown below, this figure further illustrates the impact of the spatial resolution limit of the optical system on microcirculation monitoring through a local microstructure image and a comparison of different regularization effects. In the local microstructure image on the left, extremely fine capillaries with diameters smaller than the full width at half maximum (FWHM) of the optical system's PSF (Point Spread Function) (e.g., the 1 μm reference size shown in the figure) are clearly marked. Because the physical size of these extremely fine capillaries has exceeded or is close to the resolution limit of hardware imaging, focal blood oxygenation mutations at the microscale often occur within their regions. Combined with... Figure 7 As shown in the comparison image on the lower right, if a traditional, globally uniform, excessive regularization constraint is used, the algorithm tends to misjudge the aforementioned minute, real physiological mutations as high-frequency noise and forcibly erase them. As clearly indicated in the rightmost magnified view, this leads to the loss of extremely fine capillaries and blood oxygenation mutation information, smoothing out the originally real focal blood oxygenation mutations into a blurred, continuous background. To address this limitation, this embodiment utilizes the aforementioned hardware resolution-aware variable-weight regularization mechanism. When a vascular structure is identified to be smaller than the PSF half-width, a moderate regularization constraint, as shown in the center of the image, is adaptively applied. This effectively avoids excessive smoothing of minute anatomical structures, ensuring effective denoising of the macroscopic background while preserving the morphology of extremely fine capillaries and their accompanying focal blood oxygenation mutation characteristics to the greatest extent possible.

[0185] In summary, this application addresses the limitations of current non-contact blood oxygen detection methods, such as complex equipment, sensitivity to individual skin color differences, lack of spatial resolution, and neglect of local scattering distortion. It provides a two-dimensional blood oxygen distribution imaging method based on dual-band diffuse reflection spatial mapping. By introducing an adaptive wavelength selection mechanism, local optical path difference estimation, and variational space regularization model, it effectively compensates for optical path deviation caused by light transmission in non-uniform biological tissues at the physical level.

[0186] Simulation results based on a highly realistic multilayer biomimetic tissue model demonstrate that the method described in this application effectively suppresses halo artifacts easily generated at the edges of microvessels and local ischemic lesions by the traditional global ratio method, significantly reducing the root mean square error of blood oxygen retrieval, thereby achieving high-resolution pixel-level blood oxygen spatial mapping. Simultaneously, this method effectively overcomes the interference of transmitted light attenuation caused by different melanin concentrations, exhibiting strong measurement robustness across different skin color grades. Furthermore, based on a clear understanding of the physical boundaries between computational overhead and spatial resolution, this scheme verifies that high-quality two-dimensional blood oxygen distribution information can be obtained using simplified dual-band hardware. Therefore, the technical solution of this application provides reliable methodological support for high-precision, broad-spectrum, and non-invasive physiological parameter monitoring. In some optional engineering implementation approaches, future efforts could further combine lightweight deep learning networks and other methods to optimize the computational overhead of iterative solutions, thereby accelerating the deployment and application of non-contact medical-grade vital sign monitoring devices and enabling their application in various wearable devices (such as AR glasses).

[0187] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0188] Figure 8 A structural block diagram of an example of a blood oxygen distribution imaging system based on dual-band diffuse reflection spatial mapping according to an embodiment of this application is shown.

[0189] like Figure 8 As shown, the blood oxygen distribution imaging system 800 based on dual-band diffuse reflection spatial mapping includes a dual-band image acquisition unit 810, an image processing unit 820, an optical path correction unit 830, a blood oxygen inversion unit 840, and a distribution imaging unit 850.

[0190] The dual-band image acquisition unit 810 is used to acquire a sequence of raw dual-band diffuse reflection images of the tissue region under test in two specific bands.

[0191] The image processing unit 820 is used to preprocess and segment the original dual-band diffuse reflection image sequence to obtain a normalized dual-band image and the corresponding region of interest.

[0192] The optical path correction unit 830 is used to estimate the effective optical path ratio between the two specific bands for each pixel in the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, so as to construct a local optical path correction matrix covering the region of interest.

[0193] The blood oxygen inversion unit 840 is used to calculate the preliminary blood oxygen saturation of each pixel in the region of interest based on the reflection attenuation ratio of the normalized dual-band image in the two specific bands, combined with the effective optical path ratio of the corresponding pixel in the local optical path correction matrix, using a preset dual-band inversion model. The dual-band inversion model is configured to establish a functional mapping relationship between the reflection attenuation ratio, the effective optical path ratio, the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the two specific bands, and the preliminary blood oxygen saturation based on the improved Lambert-Beer law, so as to obtain the preliminary blood oxygen saturation of the corresponding pixel.

[0194] The distribution imaging unit 850 is used to perform spatial regularization optimization on the preliminary blood oxygen saturation of all pixels in the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

[0195] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device) to perform the steps of any of the above-described blood oxygen distribution imaging methods based on dual-band diffuse reflection spatial mapping.

[0196] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described blood oxygen distribution imaging methods based on dual-band diffuse reflection spatial mapping.

[0197] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a blood oxygen distribution imaging method based on dual-band diffuse reflectance spatial mapping.

[0198] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0199] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0200] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0201] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A blood oxygen distribution imaging method based on dual-band diffuse reflectance spatial mapping, characterized in that, The method includes: Obtain a sequence of raw dual-band diffuse reflectance images of the tissue region under two specific bands; The original dual-band diffuse reflection image sequence is preprocessed and segmented to obtain a normalized dual-band image and the corresponding region of interest. For each pixel in the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, the effective optical path ratio of each pixel between the two specific bands is estimated to construct a local optical path correction matrix covering the region of interest. For each pixel within the region of interest, based on the reflection attenuation ratio of the normalized dual-band image in the two specific bands, and combined with the effective optical path ratio of the corresponding pixel in the local optical path correction matrix, the preliminary blood oxygen saturation of the pixel is calculated using a preset dual-band inversion model. The dual-band inversion model is configured to: establish a functional mapping relationship between the reflection attenuation ratio, the effective optical path ratio, the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the two specific bands, and the preliminary blood oxygen saturation, based on the improved Lambert-Beer law, in order to obtain the preliminary blood oxygen saturation of the corresponding pixel. Spatial regularization optimization is performed on the preliminary blood oxygen saturation of all pixels within the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

2. The method according to claim 1, characterized in that, Before acquiring the original dual-band diffuse reflectance image sequence of the tissue region under two specific bands, the method further includes determining the center wavelength corresponding to the two specific bands based on an adaptive wavelength selection mechanism, specifically including: The initial skin color features and tissue thickness features corresponding to the tissue region to be tested are extracted and converted into the corresponding melanin index and tissue thickness parameters, respectively. In a pre-defined candidate wavelength spectral library, the sensitivity factor corresponding to each candidate wavelength is calculated by combining the extinction coefficients of oxyhemoglobin and deoxyhemoglobin. The sensitivity factor is proportional to the absolute value of the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin at the same candidate wavelength, and inversely proportional to the standard deviation of system noise jointly affected by the melanin index and the tissue thickness parameter. Traverse all candidate wavelength combinations in the candidate wavelength spectral library and construct an optimization function with the goal of maximizing the sum of the sensitivity factors corresponding to two candidate wavelengths and the constraint that the spectral distance between the two candidate wavelengths meets the preset constraint condition. Solve the optimization function to obtain the optimal wavelength combination, and determine the two candidate wavelengths corresponding to the optimal wavelength combination as the center wavelengths corresponding to the two specific bands; wherein, the preset constraint conditions include that the spectral distance between the two candidate wavelengths is greater than or equal to the minimum spectral distance threshold used to avoid excessively small local optical path ratios, and less than or equal to the maximum spectral distance threshold used to avoid excessively large differences in tissue scattering.

3. The method according to claim 1, characterized in that, The preprocessing and region segmentation of the original dual-band diffuse reflection image sequence to obtain a normalized dual-band image and its corresponding region of interest includes: Acquire pre-captured dark field images and whiteboard reflection images acquired based on a standard reference whiteboard; For each image frame in the dual-band diffuse reflection original image sequence, flat-field correction processing is performed using the dark field image and the whiteboard reflection image to eliminate non-uniform illumination and sensor dark current interference, generating a corresponding corrected image sequence; wherein, the flat-field correction processing includes: subtracting the dark field intensity of the dark field image from the original reflection intensity of the original image frame, and dividing by the difference between the whiteboard reflection intensity and the dark field intensity of the whiteboard reflection image to obtain the corresponding normalized reflection intensity; Subpixel-level micro-motion compensation based on adjacent frame phase correlation is performed on the corrected image sequence, and the cardiac prior information of the tissue region to be tested is initially estimated from the compensated corrected image sequence; An adaptive bandpass filter that tracks the cardiac a priori information is used to extract the frequency band of the tissue pulsation signal to filter out ambient light noise and low-frequency motion artifacts, generating a spatiotemporally aligned normalized dual-band image. The channel reflectance ratio of the normalized dual-band image under the two specific bands is extracted as the skin color feature ratio, and the initial contour of the epidermal tissue is determined by combining the edge detection operator. Within the initial contour, the pixels with homogeneous optical scattering features are structurally clustered using a superpixel segmentation algorithm to extract the set of pixels containing effective physiological tissue as the region of interest, thereby obtaining the normalized dual-band image and the corresponding region of interest.

4. The method according to claim 1, characterized in that, For each pixel within the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, the effective optical path ratio of each pixel between the two specific bands is estimated to construct a local optical path correction matrix covering the region of interest, including: For each target pixel in the region of interest, the normalized reflection intensity of each neighboring pixel in the local neighborhood of the target pixel under the two specific bands is extracted, and local logarithmic transformation is performed respectively to obtain the local logarithmic reflection intensity distribution corresponding to the two specific bands. For each of the two specific bands, a spatial linear regression model is established with the Euclidean distance from each neighboring pixel in the local neighborhood to the target pixel as the independent variable and the local logarithmic reflection intensity of the corresponding neighboring pixel as the dependent variable. The slope obtained from the fitting is determined as the reflection intensity attenuation slope corresponding to the specific band. The reflection intensity attenuation slope is used to characterize the image gradient information. Based on the diffuse reflection approximation theory, the initial optical path ratio estimate of the target pixel is determined according to the ratio of the reflection intensity attenuation slopes corresponding to the two specific wavebands. The initial optical path ratio estimate of each pixel in the region of interest is subjected to iterative regularization smoothing; wherein, the iterative regularization smoothing includes: determining Gaussian spatial distance weights based on the spatial distance between pixels in the local neighborhood, determining similarity weights based on image edge features, and using the Gaussian spatial distance weights and the similarity weights to perform joint weighted filtering on the initial optical path ratio estimate; The optical path ratio value after iterative convergence is determined as the effective optical path ratio of the corresponding pixel, and a local optical path correction matrix is ​​constructed from the effective optical path ratio of each pixel in the region of interest.

5. The method according to claim 4, characterized in that, For each pixel within the region of interest, based on the reflection attenuation ratio of the normalized dual-band image in the two specific bands, and combined with the effective optical path ratio of the corresponding pixel in the local optical path correction matrix, the preliminary blood oxygen saturation of the pixel is calculated using a preset dual-band inversion model, including: For each pixel in the region of interest, the normalized reflectance intensity in the first and second specific bands of the two specific bands is extracted, and the corresponding tissue absorbance is calculated based on the negative logarithmic transformation of the normalized reflectance intensity. The ratio of the tissue absorbance corresponding to the first specific wavelength band to the tissue absorbance corresponding to the second specific wavelength band is determined as the reflection attenuation ratio of the corresponding pixel. The reflection attenuation ratio of the corresponding pixel and the effective optical path ratio of the pixel in the local optical path correction matrix are substituted into the dual-band inversion model; wherein, the dual-band inversion model is configured to establish an inversion function mapping relationship between the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the first specific band and the second specific band, and the preliminary blood oxygen saturation, based on the improved Lambert-Beer law, and the preliminary blood oxygen saturation of the corresponding pixel is solved by eliminating the total hemoglobin concentration term; The inversion function mapping relationship is configured as follows: the preliminary blood oxygen saturation is equal to the ratio of the target numerator to the target denominator; the target numerator is obtained by subtracting the extinction coefficient of deoxyhemoglobin in the first specific band from the product of the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficient of deoxyhemoglobin in the second specific band; the target denominator is obtained by subtracting the product of the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the second specific band from the product of the reflection attenuation ratio, the effective optical path ratio, and the difference between the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the second specific band; and the effective optical path ratio is the ratio of the effective optical path corresponding to the second specific band to the effective optical path corresponding to the first specific band.

6. The method according to claim 5, characterized in that, The step of performing spatial regularization optimization on the preliminary blood oxygen saturation of all pixels within the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest includes: A global variational objective function is constructed, comprising a data fidelity term and a spatial smoothing regularization term with edge-preserving weights. The data fidelity term characterizes the consistency difference between the measured reflection attenuation ratio of the corresponding pixel and the theoretical reflection attenuation ratio derived from the dual-band inversion model based on the current blood oxygen saturation and effective optical path ratio, using the sum of squared errors. The spatial smoothing regularization term applies a weighted penalty to the square of the blood oxygen saturation difference between adjacent pixels, under the combined adjustment of local texture edge-preserving weights and global regularization weights. The edge-preserving weights are dynamically determined based on the local texture gradient of the normalized dual-band image and are used to reduce excessive smoothing of tissue boundary regions. Using the initial blood oxygen saturation corresponding to each pixel as the initial iteration value, the global variational objective function is minimized and iteratively updated using the gradient descent algorithm; wherein, the blood oxygen saturation of the next iteration is obtained by updating the blood oxygen saturation of the current iteration along the negative gradient direction of the global variational objective function at the current blood oxygen saturation, combined with a preset iteration step size; When the decrease in loss of the global variational objective function is lower than the preset convergence threshold or the maximum number of iterations is reached, the iteration stops, and the final converged blood oxygen saturation result is determined as the two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

7. The method according to claim 6, characterized in that, The step of using the preliminary blood oxygen saturation corresponding to each pixel as the initial iteration value and employing a gradient descent algorithm to minimize and iteratively update the global variational objective function includes: The point spread function half-width and height of the current optical system are determined, and the normalized dual-band image is processed using a multi-scale vascular enhancement filtering algorithm to determine the local vascular structure scale corresponding to each pixel in the region of interest. Based on the scale ratio between the local vascular structure scale and the half-width at half-maximum (WHM) of the point spread function, a pixel-wise spatial weighted regularization coefficient is determined. When calculating the spatial smoothing regularization term and its partial derivative gradient in the global variational objective function, the pixel-wise spatial weighted regularization coefficient replaces the regularization weight coefficient. Specifically, for pixels whose local vascular structure scale is smaller than the WHM of the point spread function, the pixel-wise spatial weighted regularization coefficient is reduced using a monotonically decreasing mapping function based on the scale ratio. A coupled update mechanism between local optical path and tissue oxygenation status is established during the iterative update process: In the current iteration update, based on the blood oxygen saturation of the current iteration and the reflection attenuation ratio of the corresponding pixel, the effective optical path ratio is synchronously corrected in combination with the dual-band inversion model to obtain the effective optical path ratio after the current iteration update. When calculating the global variational objective function and its partial derivative gradient for the next iteration, the effective optical path ratio in the global variational objective function is replaced with the effective optical path ratio updated in the current iteration.

8. A blood oxygen distribution imaging system based on dual-band diffuse reflectance spatial mapping, characterized in that, The system includes: The dual-band image acquisition unit is used to acquire a sequence of raw dual-band diffuse reflection images of the tissue region under test in two specific bands; The image processing unit is used to preprocess and segment the original dual-band diffuse reflection image sequence to obtain a normalized dual-band image and the corresponding region of interest. An optical path correction unit is used to estimate the effective optical path ratio between two specific bands for each pixel in the region of interest, based on the image gradient information of the normalized dual-band image in the local neighborhood of the corresponding pixel, so as to construct a local optical path correction matrix covering the region of interest. The blood oxygen inversion unit is used to calculate the preliminary blood oxygen saturation of each pixel in the region of interest based on the reflection attenuation ratio of the normalized dual-band image in the two specific bands, combined with the effective optical path ratio of the corresponding pixel in the local optical path correction matrix, using a preset dual-band inversion model. The dual-band inversion model is configured to establish a functional mapping relationship between the reflection attenuation ratio, the effective optical path ratio, and the extinction coefficients of oxyhemoglobin and deoxyhemoglobin in the two specific bands and the preliminary blood oxygen saturation, based on the improved Lambert-Beer law, so as to obtain the preliminary blood oxygen saturation of the corresponding pixel. A distribution imaging unit is used to perform spatial regularization optimization on the preliminary blood oxygen saturation of all pixels in the region of interest to generate a two-dimensional blood oxygen saturation distribution map corresponding to the region of interest.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-7.

Images