Breast blood oxygen dynamic imaging method and system based on multispectral fusion

Abstract

The invention relates to the technical field of deep learning, and provides a mammary gland blood oxygen dynamic imaging method and system based on multispectral fusion, and the method comprises the steps: irradiating mammary gland tissues, and collecting mammary gland tissue reflected light image sequences under different wavelengths to obtain multispectral time sequence data; constructing an adaptive weight prediction network model, and dynamically predicting the fusion weight of each spectral wavelength at different moments in the future according to the blood oxygen change intensity and historical time sequence information at the current moment; performing multi-resolution time sequence modeling on the multi-spectral time sequence data, respectively extracting short-term, medium-term and long-term time sequence features, and performing weighted fusion on the features of different spectral wavelengths through a fusion weight to obtain fusion feature data; and performing inversion processing on the fused feature data, and calculating oxyhemoglobin saturation distribution of a continuous time sequence in real time to obtain a dynamic blood oxygen image sequence. According to the invention, the blood oxygen detection precision and the depth resolution are improved, and high-precision dynamic imaging of the blood oxygen saturation of the breast tissue is realized.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a breast blood oxygen dynamic imaging method and system based on multi-spectral fusion. Background Art

[0002] Breast oximetry dynamic imaging technology refers to a non-invasive optical imaging system that uses multispectral optical detection to monitor the temporal and spatial distribution of blood oxygen saturation in breast tissue in real time, acquiring information on vascular function, tissue metabolic activity, and pathophysiological processes. Compared with traditional mammography, ultrasound, or magnetic resonance imaging, breast oximetry dynamic imaging offers significant advantages, including real-time performance, the absence of ionizing radiation, low cost, and portability. This technology has important applications in areas such as early breast cancer screening, neovascularization assessment, chemotherapy efficacy monitoring, breast inflammation diagnosis, and breast health management. Multispectral oximetry imaging can rapidly acquire dynamic blood oxygen information at different depths within breast tissue, enabling sensitive detection of pathological features such as hypermetabolism and angiogenesis in malignant tumors, providing important functional imaging evidence for the early detection and precise diagnosis and treatment of breast diseases.

[0003] Existing technologies for breast blood oxygen imaging primarily employ traditional multispectral fusion methods based on fixed-weight linear combinations and single-time-scale analysis. While these methods achieve basic extraction and display of blood oxygen information, in practice, they inadequately consider the optical properties of the multi-layered heterogeneous structure of breast tissue, inadequately compensate for differences in tissue penetration depths between different wavelengths, and generally employ static weight allocation strategies, ignoring temporal correlation analysis of blood oxygen changes. Consequently, in high-precision scenarios such as breast disease detection, it is difficult to accurately distinguish the blood oxygen status of different tissue layers. In particular, in complex breast tissue environments, the detection sensitivity for deep lesions is low, and the ability to suppress physiological interference signals is poor, leading to artifacts and misjudgments, impacting the accuracy and reliability of clinical diagnoses. Summary of the Invention

[0004] In view of this, the present invention proposes a breast blood oxygen dynamic imaging method and system based on multi-spectral fusion, which solves the problem that the existing technology is difficult to accurately distinguish the blood oxygen status of different tissue layers in high-precision scenarios such as breast disease detection. In particular, in complex breast tissue environments, the detection sensitivity of deep lesions is low, the ability to suppress physiological interference signals is poor, and artifacts and misjudgments are easily generated, affecting the accuracy and reliability of clinical diagnosis.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for dynamic breast blood oxygen imaging based on multispectral fusion, comprising the following steps: Multiple light sources with specific wavelengths are used to illuminate breast tissue, and image sensors are used to synchronously capture a sequence of breast tissue reflected light images at different wavelengths to obtain multispectral time series data. An adaptive weight prediction network model is constructed based on the intensity of blood oxygen changes. Based on the current intensity of blood oxygen changes and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted. Performing multi-resolution time series modeling on the multispectral time series data, extracting short-term, medium-term and long-term time series features respectively, and performing weighted fusion on the features of different spectral wavelengths by the fusion weight to obtain fused feature data; The fused feature data is inverted using a blood oxygen inversion algorithm, and the blood oxygen saturation distribution of a continuous time series is calculated in real time to obtain a dynamic blood oxygen image sequence of breast tissue.

[0006] Based on the above technical solution, preferably, the adaptive weight prediction network model is constructed based on the blood oxygen change intensity, and the fusion weight of each spectral wavelength at different moments in the future is dynamically predicted according to the current blood oxygen change intensity and historical time series information, including: A layer-aware adaptive weight prediction network model is constructed based on the optical characteristics of breast tissue layers. Based on the adaptive weight prediction network model, combined with the physiological cycle compensation mechanism and individualized anatomical structure priors, the fusion weights of each spectral wavelength at different moments in the future are dynamically predicted.

[0007] On the basis of the above technical solution, preferably, the layered-aware adaptive weight prediction network model based on the optical characteristics of breast tissue layers is constructed, including: An adaptive weight prediction network model was constructed, comprising a hierarchical spectral response module, a physiological interference suppression module, and an anatomical structure perception module. The hierarchical spectral response module established a wavelength-tissue layer mapping relationship based on the differences in penetration depth of different wavelengths in breast tissue. The physiological interference suppression module identified and filtered out periodic interference components corresponding to heart rate and respiratory rate through frequency domain analysis. The anatomical structure perception module differentially modulated the spectral weights of different spatial regions based on the radial distribution characteristics of breast tissue.

[0008] On the basis of the above technical solution, preferably, the adaptive weight prediction network model is combined with the physiological cycle compensation mechanism and the individualized anatomical structure prior to dynamically predict the fusion weight of each spectral wavelength at different moments in the future, including: The individualized optical parameter correction algorithm is used to estimate the current individual's tissue optical properties in real time. Combined with the calculation of stratified blood oxygen change intensity and physiological cycle phase detection, the stratified weight coefficient of each wavelength is dynamically adjusted. The multi-spectral fusion weight is output through a spatial weighted fusion strategy based on breast anatomy priors.

[0009] On the basis of the above technical solution, preferably, the multi-resolution time series modeling is performed on the multispectral time series data, the short-term, medium-term and long-term time series features are extracted respectively, and the features of different spectral wavelengths are weightedly fused by the fusion weight to obtain fused feature data, including: Based on the physiological time scale characteristics of breast blood oxygenation changes, a hierarchical and cascaded multi-resolution time series modeling was performed on the multispectral time series data to extract short-term features corresponding to the heartbeat cycle, medium-term features corresponding to the respiratory cycle, and long-term features corresponding to the vascular regulation cycle. The fusion weights are combined with multi-resolution time series features for depth-time joint weighting through an adaptive weighted fusion algorithm based on tissue depth perception to generate fusion feature data based on breast tissue stratified blood oxygen dynamics.

[0010] Based on the above technical solution, preferably, the inversion processing of the fused feature data by a blood oxygen inversion algorithm is performed to calculate the blood oxygen saturation distribution of a continuous time series in real time to obtain a dynamic blood oxygen image sequence of breast tissue, including: A modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship is established between the fused feature data and the blood oxygen concentration of breast tissue layers. The modified Beer-Lambert inversion model is solved by using a layered blood oxygen saturation dynamic inversion algorithm to reconstruct the spatial distribution of blood oxygen saturation at each depth layer of breast tissue in real time, thereby obtaining a time-series continuous dynamic blood oxygen image sequence.

[0011] On the basis of the above technical solution, preferably, the modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship between the fused feature data and the stratified blood oxygen concentration of breast tissue is established, including: A modified Beer-Lambert inversion model was constructed, including a multilayer scattering correction module, a hemoglobin species identification module, and an individualized optical parameter adaptation module. The multilayer scattering correction module establishes a probability distribution model of photon scattering paths in breast tissue based on Monte Carlo photon transport theory. The hemoglobin species identification module separates the contributions of oxyhemoglobin, deoxyhemoglobin, and abnormal hemoglobin using a spectral decoupling algorithm. The individualized optical parameter adaptation module dynamically adjusts the optical attenuation coefficient and scattering parameters based on the density, thickness, and age characteristics of individual breast tissue.

[0012] In a second aspect, the present invention further provides a breast blood oxygen dynamic imaging system based on multi-spectral fusion, the system comprising: An image acquisition module is used to illuminate breast tissue using light sources of multiple specific wavelengths, and synchronously collect a sequence of breast tissue reflected light images at different wavelengths through an image sensor to obtain multispectral time series data; The fusion weight module is used to build an adaptive weight prediction network model based on the blood oxygen change intensity. According to the current blood oxygen change intensity and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted; A time series modeling module is used to perform multi-resolution time series modeling on multispectral time series data, extract short-term, medium-term and long-term time series features respectively, and perform weighted fusion of features of different spectral wavelengths through the fusion weight to obtain fused feature data; The dynamic imaging module is used to perform inversion processing on the fused feature data through a blood oxygen inversion algorithm, calculate the blood oxygen saturation distribution of a continuous time series in real time, and obtain a dynamic blood oxygen image sequence of breast tissue.

[0013] In a third aspect, the present invention further provides an electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps of a breast blood oxygen dynamic imaging method based on multi-spectral fusion.

[0014] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions enable a computer to implement steps of a method for dynamic breast blood oxygen imaging based on multi-spectral fusion.

[0015] The present invention provides a method and system for dynamic breast blood oxygen imaging based on multispectral fusion, which has the following beneficial effects compared with the prior art: (1) By combining multispectral illumination with synchronous acquisition, adaptive weight dynamic prediction, multi-resolution time series feature fusion, and a modified blood oxygen inversion algorithm, the fusion weight of each wavelength is adaptively adjusted according to the intensity of blood oxygen change, effectively capturing physiological change characteristics of different time scales such as heartbeat, respiration, and vascular regulation, improving blood oxygen detection accuracy and depth resolution, and achieving high-precision dynamic imaging of breast tissue blood oxygen saturation; (2) By constructing a hierarchical perception network architecture consisting of a hierarchical spectral response module, a physiological interference suppression module, and an anatomical structure perception module, an accurate wavelength-tissue layer mapping relationship is established based on the difference in penetration depth of different wavelengths in breast tissue. Physiological interference components such as heart rate and respiration are effectively identified and filtered out based on frequency domain analysis. At the same time, differential weight modulation is performed on different spatial regions based on the radial distribution characteristics of breast tissue, thereby improving the accuracy of multispectral fusion and realizing adaptive weight prediction of the complex optical properties of breast tissue. (3) By constructing a modified Beer-Lambert inversion model based on the multi-layer optical scattering characteristics of breast tissue, an accurate nonlinear mapping relationship between the fused feature data and the stratified blood oxygen concentration of breast tissue was established. The dynamic inversion algorithm of stratified blood oxygen saturation was used to achieve real-time reconstruction of the blood oxygen distribution in each depth layer, thereby improving the depth resolution capability of blood oxygen inversion and achieving high spatiotemporal resolution imaging of the blood oxygen status of breast tissue. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of a breast blood oxygen dynamic imaging method based on multi-spectral fusion according to the present invention; Figure 2 This is a performance comparison diagram of the blood oxygen change intensity sensing network of the present invention; Figure 3 This is a comparison diagram of the physiological interference suppression effect of the present invention; Figure 4 This is a dynamic change curve diagram of the multi-spectral fusion weight of the present invention; Figure 5 This is a multi-resolution temporal feature extraction effect diagram of the present invention; Figure 5 (a) is the sub-graph of the short-term heartbeat signal feature extraction effect. Figure 5 (b) is the mid-term respiratory signal feature extraction effect sub-graph, Figure 5 Middle (c) is the subgraph of the effect of long-term vascular regulation signal feature extraction; Figure 6 This is a structural diagram of a breast blood oxygen dynamic imaging system based on multi-spectral fusion according to the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a breast blood oxygen dynamic imaging method based on multi-spectral fusion, comprising the following steps: Multiple light sources with specific wavelengths are used to illuminate breast tissue, and image sensors are used to synchronously capture a sequence of breast tissue reflected light images at different wavelengths to obtain multispectral time series data. An adaptive weight prediction network model is constructed based on the intensity of blood oxygen changes. Based on the current intensity of blood oxygen changes and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted. Performing multi-resolution time series modeling on the multispectral time series data, extracting short-term, medium-term and long-term time series features respectively, and performing weighted fusion on the features of different spectral wavelengths by the fusion weight to obtain fused feature data; The fused feature data is inverted by using a blood oxygen inversion algorithm based on the modified Beer-Lambert law, and the blood oxygen saturation distribution of the continuous time series is calculated in real time to obtain a dynamic blood oxygen image sequence of breast tissue.

[0020] Specifically, this embodiment combines multi-spectral illumination with synchronous acquisition, adaptive weight dynamic prediction, multi-resolution time series feature fusion, and a modified blood oxygen inversion algorithm to adaptively adjust the fusion weight of each wavelength according to the intensity of blood oxygen change, effectively capturing physiological change characteristics at different time scales such as heartbeat, respiration, and vascular regulation, thereby improving blood oxygen detection accuracy and depth resolution, and achieving high-precision dynamic imaging of breast tissue blood oxygen saturation.

[0021] The method comprises: irradiating breast tissue with light sources of multiple specific wavelengths, synchronously collecting a sequence of images of breast tissue reflected light at different wavelengths through an image sensor, and obtaining multispectral time series data, including: Control a light source array consisting of light sources of multiple preset wavelengths to irradiate breast tissue according to a predetermined timing pattern.

[0022] In a specific embodiment, a light source array consisting of light-emitting diodes of at least seven specific wavelengths is used, wherein the specific wavelengths include 660nm, 730nm, 780nm, 810nm, 850nm, 880nm and 940nm, and the breast tissue is irradiated one by one or in combination according to a preset time-sharing drive sequence.

[0023] Through a trigger control algorithm based on clock synchronization, at the corresponding moment when each wavelength light source in the light source array irradiates the breast tissue, the image sensor is synchronously triggered to expose and collect the reflected light image of the breast tissue at the wavelength at the corresponding moment, thereby obtaining multispectral time series data.

[0024] In a specific embodiment, a high-speed CMOS image sensor is used as the image sensor. The trigger control algorithm based on clock synchronization generates a synchronous clock signal through a unified clock source, accurately controls the exposure start and stop of the high-speed CMOS image sensor and strictly synchronizes the on and off times of the light sources of different wavelengths in the light source array, and collects breast tissue reflected light images corresponding to each wavelength at an image sampling frequency of not less than 30 frames per second.

[0025] Specifically, this embodiment achieves high-quality acquisition of multi-spectral time series data by constructing a time-sharing drive timing control system for a multi-wavelength light-emitting diode array and a precise synchronization trigger mechanism based on a unified clock source. By utilizing the rational configuration of seven specific wavelengths, the entire breast tissue penetration range from shallow skin to deep pectoral muscle is covered. The time-sharing drive mode effectively avoids optical crosstalk and mutual interference between wavelengths, while the strict clock synchronization between the high-speed CMOS sensor and the light source array ensures the temporal consistency and spatial registration accuracy of the image data of each wavelength. Maintaining a high sampling frequency of not less than 30 frames per second ensures the real-time tracking capability of physiological changes such as heartbeat and respiration, provides a high-quality raw data foundation for multi-resolution time series modeling and dynamic weight prediction, improves the signal-to-noise ratio and time resolution of multi-spectral fusion imaging, and overcomes the technical limitations of traditional single-wavelength or low-frequency sampling methods in dynamic blood oxygen monitoring.

[0026] The adaptive weight prediction network model based on blood oxygen variation intensity is constructed, and the fusion weights of each spectral wavelength at different moments in the future are dynamically predicted according to the current blood oxygen variation intensity and historical time series information, including: A layer-aware adaptive weight prediction network model is constructed based on the optical characteristics of breast tissue layers. Based on the adaptive weight prediction network model, combined with the physiological cycle compensation mechanism and individualized anatomical structure priors, the fusion weights of each spectral wavelength at different moments in the future are dynamically predicted.

[0027] In one specific embodiment, the optical parameters of breast tissue layers are shown in Table 1. The optical parameters in Table 1 are derived from Monte Carlo photon propagation simulations and experimental measurements of ex vivo breast tissue samples using an integrating sphere spectrometer and a time-domain near-infrared spectroscopy system. The samples were derived from MRI-guided tissue layer thickness measurements of 30 healthy female volunteers (aged 25-45 years).

[0028] This embodiment establishes a four-layer breast tissue model based on high-resolution ultrasound and MRI imaging, which improves the depth detection accuracy compared with the traditional two-layer model; the absorption coefficient Measured in the 760nm near-infrared window, this wavelength is most sensitive to hemoglobin and is the optimal wavelength for blood oxygen detection; scattering coefficient Used to reflect the microstructural characteristics of breast tissue, the glandular tissue has the highest scattering coefficient (180±25 ), providing key features for layer identification; refractive index n This parameter describes the difference in refractive index between layers (1.35-1.41), providing a physical basis for optical boundary detection and depth positioning algorithms. The anisotropy factor, g, characterizes the forward directionality of scattering. A g value close to 1 indicates strong forward scattering, enabling deep penetration detection. By using precise layered optical parameters, this embodiment achieves a depth resolution of 3.2 mm, a 62% improvement over traditional methods.

[0029] Table 1 Optical parameters of breast tissue layers

[0030] Specifically, this embodiment achieves accurate modeling and dynamic weight optimization of the complex optical properties of breast tissue by constructing an adaptive weight prediction network model based on precise layered optical parameters. The four-layer breast tissue optical parameter table established based on Monte Carlo photon transmission simulation and large-sample experimental measurement provides a reliable physical basis for the spectral response characteristics of different depth layers, enabling the weight distribution strategy to be adaptively adjusted according to the optical differences of each tissue layer. The selection of the optimal wavelength of 760nm and the precise characterization of the scattering coefficient, refractive index, and anisotropy factor of each layer effectively solve the technical problems of fixed weights and inaccurate depth positioning in traditional methods, and significantly improves the depth resolution and blood oxygen detection accuracy. Combined with the dynamic prediction mechanism of physiological cycle compensation and individualized anatomical structure priors, this embodiment can adapt to the differences in tissue structure and physiological changes of different individuals in real time, providing intelligent weight modulation capabilities for multi-spectral fusion.

[0031] In a specific embodiment, see Figure 2 , Figure 2 This is a performance comparison chart of the blood oxygen intensity perception network for this example. The X-axis represents the method categories (traditional fixed weight fusion, single time series modeling, and layered perception), and the Y-axis represents detection accuracy (%). The three groups of data columns are labeled with a hollow cylinder and a hollow circle (traditional method), a diagonal line filled with a triangle symbol (single time series modeling), and a dot filled with a square symbol (layered perception). Figure 2It shows that the detection accuracy of the traditional fixed weight fusion method is 73.2%, the single time series modeling method is 81.6%, and the layered perception method of the present invention reaches 94.3%, which fully demonstrates the significant advantages of the layered perception adaptive weight prediction network in improving blood oxygen detection accuracy.

[0032] The layered-aware adaptive weight prediction network model based on the optical characteristics of breast tissue layers is constructed, comprising: An adaptive weight prediction network model was constructed, comprising a hierarchical spectral response module, a physiological interference suppression module, and an anatomical structure perception module. The hierarchical spectral response module established a wavelength-tissue layer mapping relationship based on the differences in penetration depth of different wavelengths in breast tissue. The physiological interference suppression module identified and filtered out periodic interference components corresponding to heart rate and respiratory rate through frequency domain analysis. The anatomical structure perception module differentially modulated the spectral weights of different spatial regions based on the radial distribution characteristics of breast tissue.

[0033] In a specific embodiment, the calculation formula of the comprehensive layered response of the layered spectral response module is: ; ; in, For the i The integrated layered response of wavelength, For the l The weight coefficient of the tissue layer, L is the total number of breast tissue layers, For the i Wavelength at l The spectral response intensity of the tissue layer, for t Moment i wavelength of incident light intensity, For the i Wavelength at l Optical attenuation coefficient of tissue layer, For the l The thickness of the tissue layer, For the l The surface reflectivity of the tissue layer, is an exponential function.

[0034] In one specific embodiment, the layered spectral response module constructs a seven-layer tissue mapping system, including the epidermis (0-0.1mm), dermis (0.1-2mm), subcutaneous fat (2-10mm), superficial glandular layer (10-20mm), deep glandular layer (20-35mm), pectoralis fascia layer (35-40mm), and pectoralis major layer (40-50mm). A penetration depth coefficient matrix is ​​established for each wavelength: 660nm primarily responds to the epidermis and dermis, 810nm and 850nm are most sensitive to the glandular layer, and 940nm can effectively penetrate the pectoralis muscle layer. Monte Carlo simulations are used to establish 10,000 photon transmission paths, constructing a distribution map of the response intensity of each wavelength in different tissue layers.

[0035] Specifically, traditional methods ignore the differences in penetration depths of different wavelengths, resulting in shallow signals masking deep information. This embodiment establishes a precise wavelength-tissue layer mapping relationship, decomposes the contribution of each wavelength according to tissue depth, and utilizes the hierarchical extension of the Beer-Lambert law, combined with tissue layer-specific optical parameters, to achieve depth positioning and hierarchical processing of signals of each wavelength. This embodiment improves the depth resolution from the traditional 5-10mm to 1-2mm, achieving millimeter-level tissue layer recognition accuracy; the signal separation of different tissue layers is increased by 65%, effectively solving the problem of signal aliasing; the effective utilization rate of each wavelength is increased by 40%, improving the detection capability of deep tissues.

[0036] In a specific embodiment, the calculation formula of the physiological interference suppression module is: ; ; ; in, After removing physiological interference i Wavelength light intensity signal, is the inverse fast Fourier transform function, is the fast Fourier transform function, for t Moment i wavelength of incident light intensity, is the heart rate frequency filter function, is the respiratory frequency filter function, is a rectangular window function, is the fundamental frequency of heart rate, is the heart rate bandwidth, is the fundamental frequency of breathing, is the respiratory bandwidth.

[0037] In one specific embodiment, the physiological interference suppression module utilizes a dual-frequency filtering system. The heart rate filter covers the 0.8-3.0 Hz frequency band (corresponding to 48-180 bpm), and the respiration filter covers the 0.1-0.5 Hz frequency band (corresponding to 6-30 beats / minute). Adaptive window functions are used, with a heart rate detection window width of 0.2 Hz and a respiration detection window width of 0.05 Hz. A real-time spectrum analysis algorithm updates individualized physiological frequency parameters 50 times per second, dynamically adjusting the filter center frequency and bandwidth. Heart rate variability analysis and irregular breathing detection are integrated to identify and compensate for abnormal physiological patterns.

[0038] Specifically, traditional fixed filtering methods cannot adapt to individual differences and changes in physiological state. This embodiment uses fast Fourier transform to analyze the frequency domain characteristics of the light intensity signal in real time, automatically identifies the heart rate and respiratory fundamental frequency, and uses a combination of adaptive bandpass and bandstop filters to accurately remove periodic physiological interference while retaining blood oxygen change information, avoiding the loss of useful signals caused by excessive filtering. This embodiment achieves a heart rate interference suppression rate of over 95% and a respiratory interference suppression rate of over 92%; the signal-to-noise ratio is improved by 8-12dB, and the purity of the blood oxygen signal is significantly improved; the adaptability to irregular heart rhythm and breathing patterns is improved by 75%.

[0039] In a specific embodiment, see Figure 3 , Figure 3 This is a comparison chart of the physiological interference suppression effect of this embodiment, where the x-axis is frequency and the y-axis is power spectral density. It includes four curves: the original signal (solid line + hollow circles), heartbeat interference (dashed line + solid dots), respiratory interference (dash-dotted line + triangle symbols), and the filtered signal (dashed line + cross). The thick dashed line in the figure marks the -25 dB suppression reference line, emphasizing that the filtered signal power is below this threshold, sufficient to verify the interference suppression capability of this embodiment.

[0040] Specifically, this embodiment achieves precise modeling and intelligent perception of the complex optical environment of breast tissue by constructing a triple collaborative network architecture consisting of a layered spectral response module, a physiological interference suppression module, and an anatomical structure perception module. The layered spectral response module, based on the precise wavelength-tissue layer mapping established by optical attenuation theory, dynamically adjusts the weight coefficients of each layer according to the penetration depth characteristics of different wavelengths, effectively resolving the depth positioning error caused by traditional methods that ignore tissue stratification. The physiological interference suppression module, through fast Fourier transform and dual filter design, accurately identifies and eliminates periodic interference components of heart rate and respiratory rate, improving the signal-to-noise ratio and detection stability of the blood oxygen signal. The anatomical structure perception module, based on the spatial heterogeneity of the radial distribution of breast tissue, implements differentiated intelligent modulation of spectral weights in different regions, effectively adapting to individual anatomical differences. The organic integration of these three modules overcomes the problems of traditional fixed-weight methods, such as irrational weight distribution, insufficient interference suppression, and poor spatial adaptability in complex physiological environments, achieving a dual improvement in depth perception and anti-interference performance.

[0041] The adaptive weight prediction network model, combined with the physiological cycle compensation mechanism and individualized anatomical structure priors, dynamically predicts the fusion weights of each spectral wavelength at different moments in the future, including: Through the individualized optical parameter correction algorithm, the current individual's tissue optical properties are estimated in real time. Combined with the stratified blood oxygen change intensity calculation and physiological cycle phase detection, the stratified weight coefficient of each wavelength is dynamically adjusted. Through the spatial weighted fusion strategy based on breast anatomy prior, the multispectral fusion weight that takes into account tissue heterogeneity is output.

[0042] In a specific embodiment, the calculation formula of the spatial weighted fusion strategy based on breast anatomy prior is: ; ; in, is the radial position r Place i The spatially weighted response of wavelength, For the i The integrated layered response of wavelength, is an exponential function, r is the radial distance relative to the center of the breast tissue, is the radial coordinate of the center of the breast tissue, is the radial position r The radial Gaussian attenuation parameter at , For the i The spatially weighted response of wavelength, is the radial position r The spatial weight coefficient at For the i Wavelength at radial position r The phase angle at .

[0043] In one specific embodiment, the anatomical structure perception module establishes a radial distribution feature model of the breast, establishes a polar coordinate system centered on the center of the breast tissue, and divides the breast area into four concentric rings: the central area (0-2cm), the inner area (2-4cm), the outer area (4-6cm), and the edge area (6-8cm). A dedicated spectral weight modulation strategy is designed for each area: the central area uses high-frequency wavelength enhancement, the outer area uses low-frequency wavelength optimization, and the edge area uses multi-wavelength coordination. Through radial Gaussian attenuation function and spatial phase angle modulation, refined differential modulation of the spectral weights of different regions is achieved.

[0044] Specifically, breast tissue has obvious radial distribution characteristics, and there are significant differences in tissue density, vascular distribution and metabolic activity in different regions. This embodiment uses a spatially weighted response function to dynamically adjust the weight coefficient of each wavelength according to the radial distance and angular position, and uses an individualized effective attenuation coefficient to compensate for tissue heterogeneity, thereby realizing a spatially adaptive spectral fusion strategy. This embodiment achieves a 60% improvement in the uniformity of detection accuracy in different breast regions, eliminating the problems of over-enhancement in the central area and insufficient signal in the edge area of ​​the traditional method; the adaptability to breast tissue heterogeneity is improved by 80%, effectively handling the complex situation of mixed distribution of dense glands and adipose tissue; the spatial resolution is increased to 0.5mm, achieving sub-millimeter-level fine imaging capabilities.

[0045] In a specific embodiment, the calculation formula of the fusion weight is: ; ; in, For the i Wavelength t The fusion weight of the moment, For the i The spatially weighted response of wavelength, is an exponential function, is the tissue heterogeneity weight factor, For the i Wavelength t The individualized effective attenuation coefficient at the moment, For the j Wavelength t The individualized effective attenuation coefficient at the moment, For the i Reference attenuation coefficient for wavelength, For the j The spatially weighted response of wavelength, For the i Individual differences in wavelength adjustment parameters, For the i Wavelength t The intensity change relative to the baseline at a given moment.

[0046] In one embodiment, see Figure 4 , Figure 4 This is a dynamic change curve of the multispectral fusion weight of this embodiment, where the X-axis represents the observation time (seconds) and the Y-axis represents the normalized fusion weight. Figure 4 Five curves are plotted, representing the weight evolution of five different wavelengths: 660 nm, 735 nm, 760 nm, 808 nm, and 850 nm. Each curve is distinguished by a different line type and symbol combination: 660 nm is a solid line with a hollow circle, 735 nm is a dashed line with a solid dot, 760 nm is a dotted line with a hollow triangle, 808 nm is a double-dash line with a hollow square, and 850 nm is a dashed line with a cross. Figure 4 The adaptive weight prediction network model is demonstrated, which dynamically adjusts the fusion weights of each wavelength in real time according to the intensity of blood oxygen changes and physiological cycles, reflecting the significant progress of the present invention over the traditional fixed weight method.

[0047] The multi-resolution time series modeling is performed on the multispectral time series data to extract short-term, medium-term and long-term time series features respectively, and the features of different spectral wavelengths are weightedly fused by the fusion weight to obtain fused feature data, including: Based on the physiological time scale characteristics of breast blood oxygen changes, a hierarchical and cascaded multi-resolution time series modeling was performed on the multispectral time series data to extract short-term features corresponding to the heartbeat cycle, medium-term features corresponding to the respiratory cycle, and long-term features corresponding to the vascular regulation cycle.

[0048] In a specific embodiment, a hierarchical cascade timing modeling network is constructed, including a short-term rapid change capture module, a medium-term periodicity recognition module, and a long-term trend prediction module. The short-term rapid change capture module uses a high-frequency sensitive timing convolutional network to extract heartbeat-related blood oxygen pulsation features within a 0.5-3 second time window. The medium-term periodicity recognition module uses a periodic attention mechanism to identify respiratory modulated blood oxygen change patterns within a 3-20 second time window. The long-term trend prediction module uses a recursive neural network to model the slow blood oxygen change trend caused by vascular autoregulation within a 20-300 second time window.

[0049] Specifically, this embodiment achieves comprehensive capture and accurate modeling of the multi-timescale physiological characteristics of breast blood oxygenation changes by constructing a hierarchical, cascaded multi-resolution temporal modeling network. The short-term rapid change capture module utilizes a high-frequency, sensitive temporal convolutional network to accurately extract rapid blood oxygenation pulsation characteristics within the heartbeat cycle, effectively capturing subtle dynamic changes within a 0.5-3 second time window. The medium-term periodicity recognition module employs a periodic attention mechanism to accurately identify 3-20 second periodic blood oxygenation change patterns caused by respiratory modulation, enhancing the ability to perceive the physiological impact of respiratory modulation. The long-term trend prediction module leverages the memory properties of a recurrent neural network to effectively model the 20-300 second slow blood oxygenation change trends caused by vascular autoregulation, enabling accurate prediction of long-term physiological regulation processes. The hierarchical, cascaded structure of these three modules overcomes the inability of traditional single-timescale modeling methods to simultaneously account for both rapid changes and long-term trends. By synergistically integrating features at multiple time scales, it improves the temporal resolution and physiological adaptability of dynamic blood oxygenation imaging, achieving full temporal coverage from rapid response at the second level to long-term monitoring at the minute level.

[0050] In a specific embodiment, see Figure 5 , Figure 5 This is the effect diagram of multi-resolution temporal feature extraction in this embodiment. Figure 5 (a) is the sub-graph of the short-term heartbeat signal feature extraction effect. Figure 5 (b) is the mid-term respiratory signal feature extraction effect sub-graph, Figure 5 (c) is a sub-graph showing the effect of extracting long-term vascular regulation signal features. Each sub-graph uses time as the horizontal axis and normalized signal amplitude as the vertical axis. The curves use different line types and symbols to distinguish feature types: short-term heartbeat is a solid line + dot, medium-term respiration is a dotted line + triangle, and long-term vascular regulation is a dot-dashed line + box. Figure 5 The technical effect of the hierarchical cascade multi-resolution timing modeling network in extracting physiological characteristics of different time scales corresponding to the heartbeat cycle, respiratory cycle and vascular regulation cycle respectively was demonstrated.

[0051] The fusion weights are combined with multi-resolution time series features for depth-time joint weighting through an adaptive weighted fusion algorithm based on tissue depth perception to generate fusion feature data based on breast tissue stratified blood oxygen dynamics.

[0052] In a specific embodiment, a depth-wavelength mapping matrix is ​​established based on the difference in penetration depth of different spectral wavelengths in breast tissue, and the multi-resolution time series features of each wavelength are grouped according to their corresponding tissue depth. The different time series features of the same depth layer are weightedly fused through a depth-adaptive weight modulation algorithm, and a cross-depth feature interaction mechanism is used to achieve the coordinated expression of blood oxygen information in different tissue layers, and finally the fused feature data containing the complete breast tissue blood oxygen dynamic information is output.

[0053] Specifically, this embodiment implements intelligent depth-time joint weighted processing of multispectral temporal features by constructing an adaptive weighted fusion algorithm based on tissue depth perception. The establishment of a depth-wavelength mapping matrix accurately locates the tissue layer corresponding to each wavelength based on the differences in penetration depth of different wavelengths. This solves the problem of insufficient positioning accuracy caused by the mixing of information from different depths in traditional methods, improving depth resolution and layer recognition capabilities. The depth-adaptive weight modulation algorithm eliminates inter-layer interference by intelligently weighting and fusing different temporal features within the same depth layer, optimizing the feature expression quality of each tissue layer and enhancing the accuracy of blood oxygen detection. The cross-depth feature interaction mechanism enables the coordinated expression and complementary enhancement of blood oxygen information from different tissue layers, such as superficial skin, mid-layer glands, and deep pectoral muscles. This overcomes the problem of information isolation in traditional layered processing and forms a complete and coherent representation of blood oxygen dynamics across the entire breast tissue. By combining depth perception and temporal modeling, this adaptive weighted fusion algorithm achieves a technological leap from single-layer detection to multi-layer collaboration, and from static analysis to dynamic tracking, improving the spatial resolution and temporal continuity of blood oxygen monitoring in complex tissue structures.

[0054] The method of performing inversion processing on the fused feature data by using a blood oxygen inversion algorithm, calculating the blood oxygen saturation distribution of a continuous time series in real time, and obtaining a dynamic blood oxygen image sequence of breast tissue includes: A modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship is established between the fused feature data and the blood oxygen concentration of breast tissue layers. The modified Beer-Lambert inversion model is solved by using a layered blood oxygen saturation dynamic inversion algorithm to reconstruct the spatial distribution of blood oxygen saturation at each depth layer of breast tissue in real time, thereby obtaining a time-series continuous dynamic blood oxygen image sequence.

[0055] Specifically, this embodiment achieves high-precision real-time conversion from fused feature data to breast tissue blood oxygen images by constructing a dynamic inversion algorithm for layered blood oxygen saturation based on a modified Beer-Lambert model. The modified Beer-Lambert inversion model fully considers the multi-layer optical scattering characteristics of breast tissue and establishes a precise nonlinear mapping relationship between fused feature data and layered blood oxygen concentration. This effectively overcomes the inversion error caused by the traditional Beer-Lambert law in multi-layer complex tissues due to the neglect of scattering effects and inter-layer interactions, thereby improving the accuracy and depth resolution of blood oxygen concentration calculation. By solving the modified model in real time, the dynamic inversion algorithm for layered blood oxygen saturation can accurately reconstruct the spatial distribution of blood oxygen saturation at different depths, such as the skin layer, fat layer, glandular layer, and pectoral muscle layer, achieving full-range blood oxygen monitoring from the surface to the deep layer, breaking through the technical bottleneck of traditional methods that are difficult to accurately detect blood oxygen in deep tissues. The real-time computing capability of the layered blood oxygen saturation dynamic inversion algorithm based on the modified Beer-Lambert model in this embodiment ensures the dynamic generation of continuous time series blood oxygen images, provides clinicians with a spatiotemporal continuous observation window for blood oxygen changes, improves the technical level of early diagnosis and dynamic monitoring of breast diseases, and achieves a technological breakthrough from static blood oxygen detection to dynamic blood oxygen imaging, providing technical support for early detection and treatment evaluation of diseases such as breast cancer.

[0056] The modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship between the fused feature data and the stratified blood oxygen concentration of breast tissue is established, including: A modified Beer-Lambert inversion model was constructed, including a multilayer scattering correction module, a hemoglobin species identification module, and an individualized optical parameter adaptation module. The multilayer scattering correction module establishes a probability distribution model of photon scattering paths in breast tissue based on Monte Carlo photon transport theory. The hemoglobin species identification module separates the contributions of oxyhemoglobin, deoxyhemoglobin, and abnormal hemoglobin using a spectral decoupling algorithm. The individualized optical parameter adaptation module dynamically adjusts the optical attenuation coefficient and scattering parameters based on the density, thickness, and age characteristics of individual breast tissue.

[0057] Specifically, this embodiment achieves accurate modeling of the complex optical environment of breast tissue and high-precision blood oxygen inversion by constructing a modified Beer-Lambert inversion model with three core modules: multi-layer scattering correction, hemoglobin species identification, and personalized optical parameter adaptation. The multi-layer scattering correction module is based on a probability distribution model established by Monte Carlo photon transmission theory, which effectively solves the fundamental defect of the traditional Beer-Lambert law that ignores the scattering effect. It improves the accuracy of optical path calculation through the scattering path length correction factor and tissue stratification weight function. The hemoglobin species identification module uses a spectral decoupling algorithm to accurately separate the contributions of different hemoglobin components, providing a reliable foundation for accurate blood oxygen calculation. The personalized optical parameter adaptation module dynamically adjusts optical parameters according to individual differences, effectively adapting to changes in tissue characteristics of different patients.

[0058] In one specific embodiment, validation data for the modified Beer-Lambert inversion model are shown in Table 2. This validation experiment employed a combination of a phantom with controllable blood oxygen saturation and a clinical control study. The phantom was prepared using a 10% fat emulsion injection and carbon ink, and blood oxygen saturation was calibrated using a dissolved oxygen meter. The clinical validation included 50 healthy volunteers, using pulse oximetry as the gold standard control.

[0059] The traditional Beer-Lambert law has certain limitations: it assumes a fixed optical path length and ignores scattering effects, resulting in large errors in complex tissues. This embodiment introduces a scattering path length correction factor, adds a tissue stratification weighting function, and utilizes a real-time optical path compensation algorithm to make corrections. This embodiment reduces the absolute error from 6.4±9.7% of the traditional method to 0.5±2.1%, a 92% improvement; controls the relative error within the full blood oxygen range to within 1%, meeting clinical accuracy requirements; reduces the standard deviation from 9.7% to 2.1%, significantly enhancing the robustness of the algorithm; and maintains high accuracy across the blood oxygen range of 52.4%-95.2%, overcoming the traditional method's problem of decreased accuracy in low blood oxygen levels.

[0060] Table 2 Validation data of the modified Beer-Lambert inversion model

[0061] The modified Beer-Lambert inversion model is solved by using a layered blood oxygen saturation dynamic inversion algorithm to reconstruct the spatial distribution of blood oxygen saturation at each depth layer of breast tissue in real time, thereby obtaining a time-series continuous dynamic blood oxygen image sequence, including: A layered iterative inversion strategy is used to gradually solve the blood oxygen saturation distribution of each tissue layer from shallow to deep. The influence of optical signal attenuation in deep tissue is corrected through a residual feedback mechanism. Combined with time domain smoothing constraints, the physiological rationality of blood oxygen changes in continuous time series is ensured. Finally, a dynamic blood oxygen image sequence of breast tissue with high spatiotemporal resolution is generated through spatial interpolation and time series registration algorithms.

[0062] Specifically, this embodiment achieves accurate reconstruction and dynamic imaging of blood oxygen saturation at multiple depths in breast tissue by constructing a layered iterative inversion strategy. This layered iterative inversion strategy employs a stepwise approach from shallow to deep layers, effectively avoiding the error accumulation and inter-layer interference issues encountered in traditional synchronous inversion. This improves depth positioning accuracy and the independence of blood oxygen saturation calculations at each layer.

[0063] The residual feedback mechanism overcomes the sharp drop in accuracy caused by signal attenuation in deep tissues, a problem often encountered in traditional deep-layer detection methods. This enables high-precision blood oxygen detection at all depths, from the surface skin to the deep pectoral muscles. The temporal smoothing constraint mechanism ensures the physiological plausibility of continuous time series blood oxygen changes, eliminates nonphysiological mutations during the inversion process, and ensures the continuity and smoothness of dynamic blood oxygen images in the temporal dimension, consistent with the physiological laws of blood oxygen variation.

[0064] See also Figure 6 The present invention also provides a breast blood oxygen dynamic imaging system based on multi-spectral fusion, the system comprising: An image acquisition module is used to illuminate breast tissue using light sources of multiple specific wavelengths, and synchronously collect a sequence of breast tissue reflected light images at different wavelengths through an image sensor to obtain multispectral time series data; The fusion weight module is used to build an adaptive weight prediction network model based on the blood oxygen change intensity. According to the current blood oxygen change intensity and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted; A time series modeling module is used to perform multi-resolution time series modeling on multispectral time series data, extract short-term, medium-term and long-term time series features respectively, and perform weighted fusion of features of different spectral wavelengths through the fusion weight to obtain fused feature data; The dynamic imaging module is used to perform inversion processing on the fused feature data through a blood oxygen inversion algorithm, calculate the blood oxygen saturation distribution of a continuous time series in real time, and obtain a dynamic blood oxygen image sequence of breast tissue.

[0065] Specifically, this embodiment of a dynamic breast blood oxygen imaging system based on multispectral fusion achieves a technological breakthrough in breast tissue blood oxygen detection, moving from static single-point imaging to dynamic, full-area imaging, through the collaborative operation of four core modules. The image acquisition module utilizes synchronized illumination from multiple specific wavelength light sources and a parallel acquisition strategy using image sensors to acquire rich multispectral time series data. The fusion weight module, through intelligent processing using an adaptive weight prediction network, dynamically optimizes the fusion weights for each spectral wavelength. It performs intelligent predictions based on real-time blood oxygen intensity changes and historical time series information, enhancing the system's adaptability and anti-interference capabilities to complex physiological environmental changes. The time series modeling module, using multi-resolution time series modeling technology, comprehensively extracts physiological features at different time scales, such as short-term heartbeat, mid-term respiration, and long-term vascular regulation. Combined with dynamic fusion weights, it intelligently integrates multi-dimensional features, overcoming the single-dimensionality and limitations of traditional methods in time series feature extraction. The dynamic imaging module, through real-time processing of a blood oxygen inversion algorithm, accurately calculates the blood oxygen saturation distribution over a continuous time series, generating dynamic blood oxygen image sequences with high temporal and spatial resolution, providing intuitive and continuous visualization of blood oxygen changes for clinical diagnosis.

[0066] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a breast blood oxygen dynamic imaging method based on multi-spectral fusion.

[0067] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the multispectral fusion-based dynamic breast oximetry imaging method described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A breast blood oxygen dynamic imaging method based on multi-spectral fusion, characterized in that: The following steps are involved: Multiple light sources with specific wavelengths are used to illuminate breast tissue, and image sensors are used to synchronously capture a sequence of breast tissue reflected light images at different wavelengths to obtain multispectral time series data. An adaptive weight prediction network model is constructed based on the intensity of blood oxygen changes. Based on the current intensity of blood oxygen changes and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted. Performing multi-resolution time series modeling on the multispectral time series data, extracting short-term, medium-term and long-term time series features respectively, and performing weighted fusion on the features of different spectral wavelengths by the fusion weight to obtain fused feature data; The fused feature data is inverted using a blood oxygen inversion algorithm, and the blood oxygen saturation distribution of a continuous time series is calculated in real time to obtain a dynamic blood oxygen image sequence of breast tissue.

2. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 1, characterized in that: The adaptive weight prediction network model based on blood oxygen variation intensity is constructed, and the fusion weights of each spectral wavelength at different moments in the future are dynamically predicted according to the current blood oxygen variation intensity and historical time series information, including: A layer-aware adaptive weight prediction network model is constructed based on the optical characteristics of breast tissue layers. Based on the adaptive weight prediction network model, combined with the physiological cycle compensation mechanism and individualized anatomical structure priors, the fusion weights of each spectral wavelength at different moments in the future are dynamically predicted.

3. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 2, characterized in that: The layered-aware adaptive weight prediction network model based on the optical characteristics of breast tissue layers is constructed, comprising: An adaptive weight prediction network model was constructed, comprising a hierarchical spectral response module, a physiological interference suppression module, and an anatomical structure perception module. The hierarchical spectral response module established a wavelength-tissue layer mapping relationship based on the differences in penetration depth of different wavelengths in breast tissue. The physiological interference suppression module identified and filtered out periodic interference components corresponding to heart rate and respiratory rate through frequency domain analysis. The anatomical structure perception module differentially modulated the spectral weights of different spatial regions based on the radial distribution characteristics of breast tissue.

4. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 3, characterized in that: The adaptive weight prediction network model, combined with the physiological cycle compensation mechanism and individualized anatomical structure priors, dynamically predicts the fusion weights of each spectral wavelength at different moments in the future, including: The individualized optical parameter correction algorithm is used to estimate the current individual's tissue optical properties in real time. Combined with the calculation of stratified blood oxygen change intensity and physiological cycle phase detection, the stratified weight coefficient of each wavelength is dynamically adjusted. The multi-spectral fusion weight is output through a spatial weighted fusion strategy based on breast anatomy priors.

5. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 1, characterized in that: The multi-resolution time series modeling is performed on the multispectral time series data to extract short-term, medium-term and long-term time series features respectively, and the features of different spectral wavelengths are weightedly fused by the fusion weight to obtain fused feature data, including: Based on the physiological time scale characteristics of breast blood oxygenation changes, a hierarchical and cascaded multi-resolution time series modeling was performed on the multispectral time series data to extract short-term features corresponding to the heartbeat cycle, medium-term features corresponding to the respiratory cycle, and long-term features corresponding to the vascular regulation cycle. The fusion weights are combined with multi-resolution time series features for depth-time joint weighting through an adaptive weighted fusion algorithm based on tissue depth perception to generate fusion feature data based on breast tissue stratified blood oxygen dynamics.

6. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 1, characterized in that: The method of performing inversion processing on the fused feature data by using a blood oxygen inversion algorithm, calculating the blood oxygen saturation distribution of a continuous time series in real time, and obtaining a dynamic blood oxygen image sequence of breast tissue includes: A modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship is established between the fused feature data and the blood oxygen concentration of breast tissue layers. The modified Beer-Lambert inversion model is solved by using a layered blood oxygen saturation dynamic inversion algorithm to reconstruct the spatial distribution of blood oxygen saturation at each depth layer of breast tissue in real time, thereby obtaining a time-series continuous dynamic blood oxygen image sequence.

7. The method for dynamic breast blood oxygen imaging based on multi-spectral fusion according to claim 6, characterized in that: The modified Beer-Lambert inversion model is constructed based on the multi-layer optical scattering characteristics of breast tissue, and a nonlinear mapping relationship between the fused feature data and the stratified blood oxygen concentration of breast tissue is established, including: A modified Beer-Lambert inversion model was constructed, including a multilayer scattering correction module, a hemoglobin species identification module, and an individualized optical parameter adaptation module. The multilayer scattering correction module establishes a probability distribution model of photon scattering paths in breast tissue based on Monte Carlo photon transport theory. The hemoglobin species identification module separates the contributions of oxyhemoglobin, deoxyhemoglobin, and abnormal hemoglobin using a spectral decoupling algorithm. The individualized optical parameter adaptation module dynamically adjusts the optical attenuation coefficient and scattering parameters based on the density, thickness, and age characteristics of individual breast tissue.

8. A breast blood oxygen dynamic imaging system based on multispectral fusion, used to execute the breast blood oxygen dynamic imaging method based on multispectral fusion according to any one of claims 1 to 7, characterized in that: The system comprises: An image acquisition module is used to illuminate breast tissue using light sources of multiple specific wavelengths, and synchronously collect a sequence of breast tissue reflected light images at different wavelengths through an image sensor to obtain multispectral time series data; The fusion weight module is used to build an adaptive weight prediction network model based on the blood oxygen change intensity. According to the current blood oxygen change intensity and historical time series information, the fusion weight of each spectral wavelength at different times in the future is dynamically predicted; A time series modeling module is used to perform multi-resolution time series modeling on multispectral time series data, extract short-term, medium-term and long-term time series features respectively, and perform weighted fusion of features of different spectral wavelengths through the fusion weight to obtain fused feature data; The dynamic imaging module is used to perform inversion processing on the fused feature data through a blood oxygen inversion algorithm, calculate the blood oxygen saturation distribution of a continuous time series in real time, and obtain a dynamic blood oxygen image sequence of breast tissue.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images