A bimodal lung ventilation imaging system and method

Abstract

The application discloses a kind of bimodal lung ventilation imaging system and method, belong to medical imaging and respiratory monitoring technical field, system includes: first light source module, emit near infrared light for lung tissue density imaging;Second light source module, emit mid-infrared light (4.24.3 μm) for CO2 concentration monitoring in lung;Array detector module is attached to chest wall surface in grid shape;Processor module, according to two waveband light signal reconstruction lung tissue density distribution and CO2 concentration distribution respectively, after spatial registration and decision level fusion, generate lung lobe level ventilation heat map.The application can distinguish invalid cavity ventilation and insufficient ventilation by bimodal fusion, realize non-invasive, continuous, lung lobe level regional ventilation function imaging, provide accurate bedside monitoring means for mechanical ventilation patient.

Description

Technical Field

[0001] This invention relates to the field of medical imaging and respiratory monitoring technology, and in particular to a dual-modal lung ventilation imaging system and method. Background Technology

[0002] Mechanical ventilation is a vital life-saving measure in intensive care, but inappropriate ventilation settings can lead to ventilator-induced lung injury (VILI). There is an urgent clinical need for a technique that can continuously, non-invasively, and bedside monitor regional lung ventilation in order to adjust ventilator parameters promptly and implement lung-protective ventilation strategies. Current lung ventilation monitoring techniques have the following limitations: 1. Pulmonary function tester: It measures airflow by using a mouthpiece, and can only obtain the total lung volume. It cannot provide information on the regional distribution of lungs, and requires the patient's cooperation. It is not suitable for comatose or sedated patients.

[0003] 2. Computed tomography (CT): It can accurately measure lung volume and regional ventilation, but it involves radiation exposure and cannot be used for continuous dynamic monitoring.

[0004] 3. Electrical impedance imaging (EIT): It can image lung ventilation distribution in real time, but its spatial resolution is limited, and it is sensitive to electrode position and the signal is easily affected by electrocardiographic interference.

[0005] 4. GASMAS-based lung monitoring technology (such as Neola Medical's products): It uses near-infrared lasers to detect oxygen and water vapor in the lungs, enabling independent monitoring of the left and right lungs, but it cannot distinguish the ventilation status of the lung lobes (upper, middle, and lower lobes).

[0006] 5. Existing carbon dioxide monitoring technologies: Mainstream or sidestream carbon dioxide monitors can only measure the CO2 concentration of exhaled air in the airway, and cannot obtain the spatial distribution of CO2 in the lungs, let alone image through the chest wall. Although some studies have used thermal imaging technology to image exhaled CO2, it monitors exhaled air rather than in situ CO2 in the lungs, and cannot penetrate the chest wall to obtain information about the lungs.

[0007] 6. Surface optical imaging technology: This method estimates lung volume by monitoring chest wall movement using distance imaging sensors. However, it can only reflect the overall movement of the chest wall and cannot distinguish the differences in local ventilation between different lobes of the lung.

[0008] Therefore, there is an urgent need to develop a non-invasive monitoring scheme that can achieve lobular ventilation imaging and provide both structural information (air volume) and functional information (ventilation efficiency), which has significant clinical and commercial value. Summary of the Invention

[0009] The purpose of this invention is to overcome the problem that the existing technology cannot achieve lobular level ventilation monitoring, and to provide a dual-modal lung ventilation imaging system and method. By simultaneously acquiring the density distribution of lung tissue (reflecting air content) and CO2 concentration distribution (reflecting ventilation efficiency), and fusing them to generate a visual ventilation heatmap, it provides an intuitive and accurate regional lung function assessment tool for clinical use.

[0010] The objective of this invention is achieved through the following technical solution: A first aspect of the present invention provides a dual-modality lung ventilation imaging system, comprising: The first light source module is used to emit first-band light that penetrates the chest wall and reflects the density of lung tissue. The second light source module is used to emit a second band of light to monitor the CO2 concentration in the lungs. The second band of light is in the mid-infrared band, and its wavelength corresponds to the absorption peak of carbon dioxide. An array-type detector module includes multiple detector units arranged in a grid pattern on the surface of the chest wall, used to receive light signals transmitted or reflected by the first light source module and the second light source module through the chest wall; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. The processor module is connected to the first light source module, the second light source module, and the array detector module, respectively, and the processor module is configured as follows: Based on the attenuation signal generated by the first band of light in the lung tissue, the lung tissue density distribution map is reconstructed; Based on the CO2 absorption signal generated in the lungs by the second band of light, a CO2 concentration distribution map in the lungs is reconstructed. By fusing the tissue density distribution map and the lung CO2 concentration distribution map, a ventilation thermogram at the lung lobe level is generated.

[0011] In some embodiments, the wavelength range of the first band light is 0.7 μm to 0.9 μm; the wavelength range of the second band light is 4.2 μm to 4.3 μm.

[0012] In some embodiments, the array detector module includes M×N detector units, where M≥3 and N≥3, and each detector unit is arranged in a grid on a flexible substrate.

[0013] In some embodiments, the reconstructed lung tissue density distribution map includes: A forward model is established to illustrate the propagation of light in a multi-layered tissue including skin, fat, muscle, bone, and lungs, where the absorption and scattering coefficients of lung tissue are dynamically adjusted according to lung ventilation status. The lung tissue density distribution map is retrieved from the attenuated signal using an iterative reconstruction algorithm.

[0014] In some embodiments, the reconstruction of the intrapulmonary CO2 concentration distribution map includes: By using differential absorption spectroscopy, the signal of the second-band light is differentiated from that of the reference wavelength to eliminate tissue background absorption and extract CO2 absorption signals related to CO2 concentration. Based on the aforementioned positive model, an iterative reconstruction algorithm is used to invert the distribution map of CO2 concentration in the lungs from the CO2 absorption signal.

[0015] In some embodiments, fusing the tissue density distribution map and the intrapulmonary CO2 concentration distribution map to generate a lobar-level ventilation thermogram includes: Spatial registration was performed between the tissue density distribution map and the lung CO2 concentration distribution map. The registered tissue density distribution map and lung CO2 concentration distribution map are fused at the pixel level. Tissue density is used as the structural layer and lung CO2 concentration is used as the functional layer. Through color mapping and overlay, a pseudo-color ventilation thermogram showing the ventilation status of different lung lobes is generated.

[0016] In some embodiments, spatial registration of the tissue density distribution map and the lung CO2 concentration distribution map specifically includes: Initial rigid alignment: Initial rigid alignment is performed based on the physical coordinates of the detector array modules; Deformable registration: A deep learning registration network based on the U-Net architecture is used to predict the three-dimensional deformation vector field from the tissue density distribution map to the lung CO2 concentration distribution map, and a spatial transformation network is applied to the tissue density distribution map to achieve spatial alignment.

[0017] In some embodiments, the pixel-level fusion of the registered tissue density distribution map and the lung CO2 concentration distribution map includes: The registered tissue density distribution map and lung CO2 concentration distribution map are input into a preset deep learning fusion network, which includes a dual encoder, a fusion module and a single decoder connected in sequence. The tissue density distribution map and the lung CO2 concentration distribution map are respectively input into two independent branches of the dual encoder to extract the tissue density distribution feature map and the lung CO2 concentration feature map respectively; The fusion module is used to stitch the tissue density distribution feature map and the lung CO2 concentration distribution feature map in the channel dimension. The channel is recalibrated through the attention mechanism, and then the channel weights are output after passing through two fully connected layers. The single decoder uses skip connections to fuse tissue density distribution feature maps and lung CO2 concentration distribution feature maps, outputting a three-channel heat map.

[0018] In some embodiments, the loss function of the deep learning fusion network is: Where L1 represents pixel-level loss, Indicates perceived loss. Let λ1, λ2, and λ3 represent the structural similarity loss, and let λ1, λ2, and λ3 represent the weighting coefficients.

[0019] A second aspect of the present invention provides a dual-modality lung ventilation imaging method, comprising the following steps: S1. Control the first light source to emit light of the first wavelength to illuminate the target's chest cavity area; S2. Control the second light source to emit second-wavelength light to illuminate the target's chest cavity area; S3. Receive the light signals transmitted or reflected by the chest wall from the first light source module and the second light source module through the array detector module; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. S4. The processor module reconstructs the lung tissue density distribution map based on the attenuation signal generated by the first band light in the lung tissue, and reconstructs the lung CO2 concentration distribution map based on the CO2 absorption signal generated by the second band light in the lung. S5. The processor module integrates the tissue density distribution map and the lung CO2 concentration distribution map to generate a ventilation thermogram at the lung lobe level.

[0020] It should be further noted that the technical features corresponding to the above-mentioned options and embodiments can be combined or substituted with each other to form new technical solutions without conflict.

[0021] Compared with the prior art, the beneficial effects of the present invention are: 1. Achieving Lung Lobe-Level Spatial Resolution: This invention employs an array-type detector module combined with a multi-layer tissue light propagation model and iterative reconstruction algorithm established in the processor module. It also introduces prior anatomical constraints on lung lobes, enabling the invention to distinguish the independent ventilation status of the upper, middle, and lower lobes. Compared to existing technologies that can only distinguish between the left and right lungs or only provide overall volume (pulmonary function testing), this invention significantly improves spatial resolution, providing more refined regional ventilation assessment for clinical practice.

[0022] 2. Dual-modal information complementarity distinguishes between dead space ventilation and inadequate ventilation: This invention simultaneously acquires density distribution maps (reflecting physical gas content) and CO2 concentration distribution maps (reflecting ventilation efficiency), and fuses them through a decision-level fusion rule function. This rule ensures that: when density is normal (low to medium value) but CO2 concentration is low (low value), the ventilation score is at a moderate level and is separately identified, corresponding to "dead space ventilation" (ventilation present but no gas exchange); when density is high (high value) and CO2 is low (low value), the ventilation score is extremely low, corresponding to "inadequate ventilation" or "atelectasis." Existing single-modal technologies cannot distinguish between these two clinical states; this invention provides a more comprehensive diagnostic basis for clinicians.

[0023] 3. Achieving Non-invasive Continuous Monitoring: This invention utilizes near-infrared light (0.7μm-0.9μm) and mid-infrared light (4.2-4.3μm) through the chest wall for detection, requiring no intubation and producing no radiation, enabling long-term bedside continuous monitoring. Furthermore, differential absorption spectroscopy technology eliminates tissue background interference, improving signal stability and anti-interference capabilities, making it suitable for scenarios requiring continuous monitoring, such as ICUs and post-anesthesia recovery rooms. Compared to CT scans, which cannot provide continuous monitoring, and EIT signals, which are susceptible to interference, this invention offers significant advantages.

[0024] 4. High-precision heterogeneous image registration to ensure fusion quality: This invention adopts a two-stage registration strategy in the processor module, which first uses rigid alignment and then deep learning deformable registration, and uses normalized mutual information as the loss function. This effectively solves the problem of intensity distribution difference caused by the different physical meanings of density map and CO2 concentration map, so that the images of the two modalities can be accurately aligned at the pixel level, thereby ensuring the clinical reliability of the subsequent fusion heat map. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a dual-modality lung ventilation imaging system according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the coverage of the left and right lung regions by a 4×4 detector array according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the detector being attached to a flexible substrate and adapted to the curved surface of the thoracic cavity, as shown in an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the propagation path of light in multiple layers of tissue (skin, fat, muscle, ribs, lungs) according to an embodiment of the present invention. Figure 5 This is a schematic diagram of lung tissue density distribution reconstruction shown in an embodiment of the present invention; Figure 6 This is a schematic diagram of the reconstructed CO2 concentration distribution map in the lungs, as shown in an embodiment of the present invention. Figure 7 The following is a ventilation thermogram of the fused lung lobe, as shown in an embodiment of the present invention. Figure 8 This is a comparison diagram showing the effect of the present invention with the prior art (left and right lung monitoring). Detailed Implementation

[0026] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that the defects in the solutions in the prior art are all the results of the inventors' practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the embodiments of this application in the following text should be the inventors' contributions to this application in the process of invention and creation, and should not be understood as technical content known to those skilled in the art.

[0028] In view of the technical problems pointed out in the background art, the present invention provides the following embodiments: In one exemplary embodiment, a dual-modal lung ventilation imaging system is provided, such as Figure 1 As shown, it includes: The first light source module is used to emit first-band light that penetrates the chest wall and reflects the density of lung tissue. The second light source module is used to emit a second band of light to monitor the CO2 concentration in the lungs. The second band of light is in the mid-infrared band, and its wavelength corresponds to the absorption peak of carbon dioxide. An array-type detector module includes multiple detector units arranged in a grid pattern on the surface of the chest wall, used to receive light signals transmitted or reflected by the first light source module and the second light source module through the chest wall; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. The processor module is connected to the first light source module, the second light source module, and the array detector module, respectively, and the processor module is configured as follows: Based on the attenuation signal generated by the first band of light in the lung tissue, the lung tissue density distribution map is reconstructed; Based on the CO2 absorption signal generated in the lungs by the second band of light, a CO2 concentration distribution map in the lungs is reconstructed. By fusing the tissue density distribution map and the lung CO2 concentration distribution map, a ventilation thermogram at the lung lobe level is generated.

[0029] Specifically, the first light source module emits light in a first wavelength band for lung tissue attenuation imaging. In this embodiment, near-infrared light (wavelength 0.7 μm to 0.9 μm) is preferred. This wavelength band has weak absorption by tissue water and hemoglobin, and its penetration depth can reach several centimeters, effectively penetrating the chest wall and reaching the lung tissue. The degree of attenuation is related to tissue density (air content): during inhalation, the lung tissue inflates, its density decreases, and light attenuation decreases; during exhalation, the lung tissue density increases, and light attenuation increases. The second light source module emits light in a second wavelength band for monitoring intrapulmonary CO2 concentration. In this embodiment, the mid-infrared band is used, with a center wavelength of 4.2-4.3 μm, corresponding to the strong absorption peak of CO2. This wavelength band is highly selective for CO2 molecules and can detect the absorption signal of intrapulmonary CO2 through the chest wall.

[0030] The array-type detector module comprises multiple photodetector units arranged in an M×N grid (M≥3, N≥3) on the chest wall surface to form a multi-channel receiving array, such as... Figure 2 As shown, the 4×4 detector array covers the left and right lung regions; the detectors can be attached to a flexible substrate to adapt to the curvature of the chest (e.g., Figure 3 As shown in the figure, a flexible patch detector is formed to receive light signals transmitted or reflected through the chest wall.

[0031] The processor module performs the following functions: Control the first and second light source modules to emit light sources in a time-division multiplexing or synchronously (frequency-division multiplexing) manner to avoid signal crosstalk; Acquire multi-channel light intensity data from the array detector module; Based on the attenuation signal of the first band of light, the density distribution map of lung tissue is reconstructed; Based on the absorption signal of the second band of light, the CO2 concentration distribution map in the lungs is reconstructed; The two images are registered and fused to generate a ventilation thermogram at the lobe level, which is then displayed on a display terminal.

[0032] Furthermore, the reconstructed lung tissue density distribution map includes: A forward model of light propagation in multi-layered tissues including skin, fat, muscle, bone, and lungs can be established using diffusion equations or Monte Carlo methods, such as... Figure 4 As shown, D1-D4 represent the photons received by each detector unit. The absorption coefficient and reduced scattering coefficient of each tissue layer are set with reference to the published tissue optical parameter database. Among them, the absorption coefficient and scattering coefficient of lung tissue are dynamically adjusted according to the lung ventilation status (exhalation / inhalation). The rib region is modeled separately as a high scattering / high absorption region in the model to avoid the interference of bony structures on lung signals.

[0033] Preferably, in this embodiment, only the absorption coefficient change related to lung ventilation is reconstructed. The scattering coefficient is fixed as a known prior parameter to reduce the degrees of freedom in reconstruction. The multipath attenuation data obtained from the probe array is substituted into the reconstruction algorithm, such as algebraic reconstruction technique (ART), conjugate gradient method, or deep learning image reconstruction network. Using an iterative reconstruction algorithm, the intrapulmonary attenuation coefficient distribution is inverted from the attenuation signal. The intrapulmonary attenuation coefficient distribution is negatively correlated with the lung tissue density distribution (air content). The reconstructed lung tissue density distribution map is shown below. Figure 5 As shown.

[0034] Furthermore, the reconstructed lung CO2 concentration distribution map includes: Differential absorption spectroscopy is used to differentiate the signal from the second wavelength (located at the CO2 absorption peak) and the reference wavelength (without absorption) to eliminate tissue background absorption and extract the CO2 absorption signal related to CO2 concentration. Based on the aforementioned forward model, a similar iterative reconstruction algorithm is used to reconstruct the intrapulmonary CO2 concentration distribution map (three-dimensional distribution) from the CO2 absorption signal. Figure 6 As shown.

[0035] Furthermore, the fusion of the tissue density distribution map and the lung CO2 concentration distribution map to generate a lung lobe-level ventilation thermogram includes: Spatial registration of the tissue density distribution map and the lung CO2 concentration distribution map (based on chest wall markers or anatomical priors) can be performed using multimodal deformable image registration technology. Based on clinical needs, fusion rules are designed. In this embodiment, the registered tissue density distribution map and the intrapulmonary CO2 concentration distribution map are fused at the pixel level. Tissue density is used as the structural layer, and intrapulmonary CO2 concentration as the functional layer. Through color mapping and overlay, a pseudo-color ventilation heatmap is generated, displaying the ventilation status of different lung lobes (hyperventilation zone, hypoventilation zone, no ventilation zone, etc.), visually showing the ventilation status of different lung lobes. Figure 7 As shown.

[0036] The tissue density distribution map and the CO2 concentration distribution map originate from multi-channel optical signals acquired by the same array detector, but the reconstruction process targets different physical quantities. Furthermore, lung tissue deformation caused by respiration leads to a nonlinear spatial misalignment between the two images, necessitating deformable registration. For example, the spatial registration of the tissue density distribution map and the intrapulmonary CO2 concentration distribution map specifically includes: Initial rigid alignment: Initial rigid alignment is performed based on the physical coordinates of the chest wall surface detector array module, using the Iterative Closest Point (ICP) algorithm; Deformable Registration: A deep learning registration network based on the U-Net architecture is used to predict the 3D deformation vector field from the tissue density distribution map to the lung CO2 concentration distribution map. A spatial transformation network is then applied to the tissue density distribution map to achieve spatial alignment. The deep learning registration network is designed as a 3-layer encoder-decoder structure. The input is a pair of dual-channel images of the tissue density distribution map and the lung CO2 concentration distribution map. The encoder uses a 3×3 convolutional kernel with a stride of 2 to downsample and extract multi-scale features. The decoder uses transposed convolutions to upsample and fuse the skip connection features of the corresponding layers of the encoder. The output is a 3D deformation vector field.

[0037] Anatomical constraints: The anatomical boundary of the lung lobe (the average lung lobe boundary template obtained through CT prior) is used as the registration constraint. An anatomical structure loss term is introduced to impose regional constraints on the reconstruction results. The regularization weight of the lung lobe boundary region is reduced, allowing for a greater degree of signal variation, thereby improving the reconstruction accuracy of the lung lobe boundary.

[0038] Specifically, the lobar boundaries (interlobar fissures) are the natural anatomical boundaries between lung lobes. In lung imaging, adjacent lobes undergo relative slippage during respiration; therefore, the reconstructed density or concentration distribution should allow for discontinuities (gradient abrupt changes) at the lobar boundaries, while maintaining smoothness within the lobes. Based on this anatomical prior, by constructing a lobar anatomical mask and introducing spatially varying regularization weights into the objective function, adaptive regularization with "weak boundary constraints and strong interior constraints" is achieved.

[0039] Lung lobe anatomical mask M ( x Construction of ) M ( x ) =exp(-d 2 ( x ) / 2σ 2 ) x Represents the spatial coordinates of any voxel within the lung. x =( x 1, x 2, x 3)∈Ω⊂R 3 Ω represents the lung region, R 3 For the three-dimensional space d ( x ) represents voxels x The Euclidean distance to the nearest lobe boundary (interlobar fissure), in mm. σ This represents the boundary width control parameter, with a recommended value of 3~5 mm, used to define the width of the transition zone around the interleaf cleft; M ( x) represents the lung lobe boundary mask, with a value range of [0,1]. It is close to 1 at the lung lobe boundary and close to 0 inside the lung lobe away from the boundary.

[0040] In image reconstruction, the objective function incorporating lung lobe anatomical constraints is:

[0041] Adaptive regularization weights W ( x Defined as: , μ ( x () represents the parameters of the lung tissue to be reconstructed, which can be the absorption coefficient. μ a (Unit: mm) -1 The target variable for reconstruction is either CO2 concentration (in % or mmHg) or CO2 concentration. This indicates the solution obtained through optimization. μ The estimated value, i.e. the final reconstruction result; A The forward propagation matrix (also known as the system matrix) describes the propagation process of light from the source to the detector, and its elements are... A ij The contribution of the j-th voxel to the sensitivity of the i-th detector channel is calculated by the optical transmission model (diffusion equation or Monte Carlo simulation); b This represents a measurement data vector, whose elements are the light intensity values ​​(units: mW or photon counts) received by each detector channel. This represents the data fitting term, which measures the reconstruction parameters. μ The difference between the generated predicted signal and the actual measured signal; λ This represents the global regularization parameter (scalar), used to balance the weights between the data fitting term and the regularization term. The recommended value range is 0.001 to 0.01. W ( x ) represents a spatially adaptive weighting function, which takes a high value inside the lung lobe (strong smoothing) and a low value at the lung lobe boundary (weak smoothing / allowing mutation), and is dimensionless; : μ The spatial gradient, i.e. The unit depends on μ Units; : Spatial adaptive regularization term, is for The squared L2 norm is used to constrain the smoothness of the reconstruction results; the weights are reduced at the boundaries, allowing... μ Gradient mutation; w 0: Basic regularization weight (scalar), used to set the smoothing intensity of the internal region of the lung lobe, recommended value range is 0.01~0.1;α This represents the boundary relaxation coefficient (scalar), with a value ranging from [0,1], and a recommended value of 0.7 to 0.9. It is used to control the proportion by which the regularization weights are reduced at the boundary, with α = 0.7. α =0.7 indicates that the weight at the boundary is reduced to 30% of that at the interior.

[0042] Furthermore, a multi-resolution progressive registration strategy is used: first, coarse registration is performed at low resolution, and then gradually refined to fine registration at high resolution to avoid registration failure caused by large deformation.

[0043] Furthermore, since the density map and CO2 concentration map have completely different intensity distributions, this invention introduces a modality-independent similarity measure—using normalized mutual information (NMI) as the registration loss function, which is unaffected by differences in image intensity distribution. Simultaneously, it evaluates the consistency of the registered image in two dimensions: anatomical structure boundaries and internal functional regions, ensuring that the registration result maintains both anatomical structure alignment and the spatial correspondence of functional information.

[0044] For example, in the registration of the density map and the CO2 concentration map, the total loss function is:

[0045] Image similarity terms L sim (using Normalized Mutual Information, NMI): Smoothing regularization terms (Deformation field smoothing constraint):

[0046] Lung lobe boundary constraint :

[0047] L total This represents the total registration loss, a dimensionless scalar quantity used to evaluate registration quality and guide network parameter updates. I fix This represents a fixed image, namely a CO2 concentration distribution map, which serves as the target reference image for registration. I mov This represents a floating image, i.e., a density distribution map, obtained through a deformation field. Perform spatial transformation to match ； Represents the deformation vector field, indicating the voxel... x A displacement vector that maps from a floating image space to a fixed image space. The unit is mm; Indicates to Apply deformation The image after sampling voxels on the floating image The gray value at the location; NMI represents normalized mutual information, with a value range of [0,1], used to measure the statistical correlation between two images, the larger the value, the more similar the images are; Represents image similarity loss. This makes the optimization direction maximize NMI (i.e. minimize the negative value). λ 1 represents the weight coefficient of the smoothing regularization term, and 0.1 to 1.0 is recommended. It is used to constrain the smoothness of the deformation field and prevent unnatural deformation. This represents the smoothness loss of the deformation field, which is achieved by summing the squares of the spatial gradient magnitudes of the deformation field to make the overall deformation field smooth. λ 2 represents the lung lobe boundary constraint weight, with a recommended value of 0.1 to 0.5, used to control the contribution of anatomical boundary information in registration; |Ω| represents the total number of voxels within the lung region Ω.

[0048] Furthermore, this invention provides a method for evaluating the consistency of registered images. It quantitatively evaluates the registration results of density maps and CO2 concentration maps from two dimensions: anatomical structure boundaries and internal functional regions. This includes: Anatomical boundary consistency assessment: First, based on the prior anatomical template of the lung lobes, binary segmentation masks for each lung lobe were extracted from both the fixed image (CO2 concentration map) and the deformed floating image (density map). The Dice similarity coefficient (DSC) was used to evaluate the overlap between the two masks; a DSC closer to 1 indicates better alignment of the lung lobe boundaries. The 95% Hausdorff distance (HD95) was used to evaluate the maximum deviation of the boundaries, in mm; a smaller value indicates higher boundary alignment accuracy. In addition, anatomical landmarks such as the intersection of the interlobar fissure and the pleura were selected, and the target registration error (TRE) of the registered landmarks was calculated.

[0049] Internal functional region consistency assessment: Normalized mutual information (NMI) is used to measure the statistical correlation of the overall intensity distribution of two images. The larger the NMI value, the more reasonable the spatial correspondence of functional information. Root mean square error (RMSE) is used to assess pixel-level intensity differences. Structural similarity index (SSIM) is used to comprehensively assess image similarity from three aspects: brightness, contrast and structure.

[0050] Verification of the physical rationality of the deformation field: Calculate the inverse consistency error (ICE) to assess the reversibility of the deformation field; the smaller the ICE, the more physically rational the deformation. Calculate the Jacobian determinant of the deformation field to ensure the reversibility of the deformation field across the entire lung region. To avoid tissue folding or tearing.

[0051] After registration, the tissue density distribution map and the CO2 concentration distribution map are in the same spatial coordinate system, allowing for pixel-level fusion.

[0052] The goal of the fusion is to output a "ventilation heatmap," where each pixel represents a comprehensive ventilation score (range 0-100) at that location. A higher score indicates better ventilation function. Let D(i,j) represent the normalized value of the tissue density distribution map (range 0-1, where 1 represents maximum density / minimum air content and 0 represents minimum density / maximum air content), and C(i,j) represent the normalized value of the CO2 concentration distribution map (range 0-1, where 1 represents normal physiological concentration and 0 represents CO2 deficiency). V(i,j) represents the final ventilation score. In this embodiment, the ventilation score determination function is as follows:

[0053] in, θ is the low CO2 concentration threshold (set to 0.3 in this embodiment, i.e., triggering the "ineffective cavity ventilation" judgment when it is 30% lower than the normal concentration). This is the indicator function. The specific state scores and state classifications are shown in Table 1.

[0054] Table 1. Illustration of State Scoring and State Classification

[0055] Furthermore, the deep learning fusion network includes a dual encoder, a fusion module, and a single decoder connected in sequence. The registered tissue density distribution map and the lung CO2 concentration distribution map are input into the preset deep learning fusion network, and the network processing procedure is as follows: Network input: Registered tissue density distribution map and lung CO2 concentration distribution map (both are single-channel grayscale images of 128×128×1).

[0056] Network structure: Dual encoder: Two independent branches, each branch contains 4 convolutional blocks, each convolutional block contains a convolutional layer (3×3, stride 2) + batch normalization + ReLU activation, progressively downsampling the image to an 8×8×256 feature map, and extracting density features and CO2 features respectively.

[0057] Fusion module: The output feature maps of the two encoders are concatenated along the channel dimension (8×8×512), the channels are recalibrated through the SE attention module (Squeeze-and-Excitation), and then the channel weights are output through two fully connected layers.

[0058] Decoder: 4 layers of transposed convolution (4×4, stride 2), progressively upsampled to 128×128×64→128×128×32→128×128×16→128×128×3, using skip connections to fuse features from corresponding layers of the dual encoders, and finally outputting a three-channel heatmap (RGB).

[0059] Furthermore, the loss function for deep learning fusion networks is:

[0060] Where L1 represents pixel-level loss, ; The perceptual loss is represented by the feature map extracted from the pre-trained VGG16 network to calculate the feature space difference. Represents structural similarity loss. λ1, λ2, and λ3 represent weighting coefficients. In this embodiment, λ1 = 0.4, λ2 = 0.3, and λ3 = 0.3 (all are empirically optimized values).

[0061] The network training strategy is as follows: Training data: A paired dataset (density map + CO2 concentration map + ventilation score heatmap) generated through Monte Carlo simulation, simulating optical measurement data under different lung lobe ventilation states. The simulation dataset consists of 10,000 sets, with 70% used for training, 15% for validation, and 15% for testing. The training data is publicly available.

[0062] Optimizer: Adam, initial learning rate It decays by 0.5 times every 50 epochs.

[0063] Training epochs: 200 epochs, batch size: 16.

[0064] like Figure 8 As shown in Table 2, the advantages of the present invention compared with existing methods are shown in Table 2.

[0065] Table 2 Comparison of the effects of the present invention and the prior art

[0066] In another exemplary embodiment, based on the same inventive concept as the method embodiment, a dual-modal lung ventilation imaging method is provided, comprising the following steps: S1. Control the first light source to emit light of the first wavelength to illuminate the target's chest cavity area; S2. Control the second light source to emit second-wavelength light to illuminate the target's chest cavity area; S3. Receive the light signals transmitted or reflected by the chest wall from the first light source module and the second light source module through the array detector module; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. S4. The processor module reconstructs the lung tissue density distribution map based on the attenuation signal generated by the first band light in the lung tissue, and reconstructs the lung CO2 concentration distribution map based on the CO2 absorption signal generated by the second band light in the lung. S5. The processor module integrates the tissue density distribution map and the lung CO2 concentration distribution map to generate a ventilation thermogram at the lung lobe level.

[0067] For example, a neonatal lung ventilation monitoring system is provided below.

[0068] Target audience: Premature infants and infants in the neonatal intensive care unit (NICU). Newborns have thin chest walls, allowing light signals to penetrate easily, making them the most suitable initial users for this invention.

[0069] Light source parameters: Primary light source: laser diode, wavelength 780 nm, power 2-5 mW, safety level conforms to IEC 60825-1 Class 1.

[0070] The second light source is a quantum cascade laser (QCL) with a wavelength of 4.26 μm and a power of 1-2 mW. The mid-infrared wavelength is located at the strong absorption peak of CO2 (the 3-5 μm band has been verified to be used for CO2 imaging), and the power is controlled within safe limits.

[0071] Detector array: 16-channel (4×4) flexible patch, with detector elements being a hybrid array of InGaAs (near-infrared) and HgCdTe (mid-infrared), or using time-division multiplexing switching filters.

[0072] Reconstruction algorithm: The U-Net deep learning network was used, and the training data came from optical simulation (tissue model built based on neonatal CT data) and a small number of clinical samples (validated by synchronous EIT).

[0073] Workflow: 1. Apply the flexible patch to the newborn's sternum and both sides, ensuring good contact; 2. Turn on the first and second light sources sequentially and collect multi-channel light intensity data; 3. The processor reconstructs density maps and CO2 concentration maps in real time (10 frames per second). 4. The data is fused to generate a ventilation heat map, displayed on the monitor, and the trend data is stored; 5. When the CO2 concentration in a lung lobe continues to decrease or the density increases, the system issues an early warning, indicating that atelectasis may occur.

[0074] Expected results: In clinical trials, this system should be able to clearly distinguish the ventilation differences between the left and right lungs, and between the upper and lower lobes of newborns, and show good consistency with EIT monitoring results.

[0075] For example, an adult ICU bedside monitoring system is provided below.

[0076] Suitable for: Adult patients with thicker chest walls (up to 5-8 cm), requiring increased light source power and detector sensitivity.

[0077] Light source parameters: First light source: multi-wavelength LED array (750 nm, 810 nm, 850 nm), average power 10-20 mW (pulse modulation to reduce average power).

[0078] Second light source: mid-infrared superluminescent light-emitting diode (SLED) or interband cascaded laser (ICL), with a wavelength adjustable from 4.2 to 4.3 μm and a peak power of 5 to 10 mW.

[0079] Detector array: 64-channel (8×8) flexible array, using avalanche photodiodes (APDs) to enhance sensitivity, and in conjunction with lock-in amplifiers to extract weak signals.

[0080] Reconstruction Algorithm: A model-based optical tomography (DOT) algorithm, combined with anatomical priors (obtained through ultrasound to assess chest wall thickness distribution), was used to improve reconstruction accuracy. Wavelength modulation spectroscopy (WMS) technology was introduced to reconstruct CO2 concentration, along with second harmonic detection, to suppress background noise.

[0081] Workflow: 1. The thickness of the patient's chest wall at various points was measured by ultrasound and used as prior information to input into the reconstruction model; 2. Time-division multiplexing of multiple light sources, synchronous acquisition by detectors; 3. Iterative reconstruction algorithm (updating one frame every 30 seconds), balancing real-time performance and image quality; 4. Ventilation thermograms are overlaid on a 3D thoracic model to visually display the ventilation distribution of the lung lobes; 5. Long-term trend analysis can assist clinical decision-making, such as monitoring the improvement of lower lobe ventilation during prone ventilation.

[0082] Safety considerations: Mid-infrared skin irradiation requires strict power density control to comply with ANSI Z136.1 standards; pulse mode is used to reduce thermal effects.

[0083] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. A dual-modal lung ventilation imaging system, characterized in that, include: The first light source module is used to emit first-band light that penetrates the chest wall and reflects the density of lung tissue. The second light source module is used to emit a second band of light to monitor the CO2 concentration in the lungs. The second band of light is in the mid-infrared band, and its wavelength corresponds to the absorption peak of carbon dioxide. An array-type detector module includes multiple detector units arranged in a grid pattern on the surface of the chest wall, used to receive light signals transmitted or reflected by the first light source module and the second light source module through the chest wall; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. The processor module is connected to the first light source module, the second light source module, and the array detector module, respectively, and the processor module is configured as follows: Based on the attenuation signal generated by the first band of light in the lung tissue, the lung tissue density distribution map is reconstructed. Based on the CO2 absorption signal generated in the lungs by the second band of light, a CO2 concentration distribution map in the lungs is reconstructed. By fusing the tissue density distribution map and the lung CO2 concentration distribution map, a ventilation thermogram at the lung lobe level is generated.

2. The dual-modality lung ventilation imaging system according to claim 1, characterized in that, The wavelength range of the first band light is 0.7 μm to 0.9 μm; the wavelength range of the second band light is 4.2 μm to 4.3 μm.

3. The dual-modality lung ventilation imaging system according to claim 1, characterized in that, The array-type detector module includes M×N detector units, where M≥3 and N≥3, and each detector unit is arranged in a grid on a flexible substrate.

4. The dual-modality lung ventilation imaging system according to claim 1, characterized in that, The reconstructed lung tissue density distribution map includes: A forward model is established to illustrate the propagation of light in a multi-layered tissue including skin, fat, muscle, bone, and lungs, where the absorption and scattering coefficients of lung tissue are dynamically adjusted according to lung ventilation status. The lung tissue density distribution map is retrieved from the attenuated signal using an iterative reconstruction algorithm.

5. A dual-modality lung ventilation imaging system according to claim 4, characterized in that, The reconstructed lung CO2 concentration distribution map includes: By using differential absorption spectroscopy, the signal of the second-band light is differentiated from that of the reference wavelength to eliminate tissue background absorption and extract CO2 absorption signals related to CO2 concentration. Based on the aforementioned positive model, an iterative reconstruction algorithm is used to invert the distribution map of CO2 concentration in the lungs from the CO2 absorption signal.

6. The dual-modality lung ventilation imaging system according to claim 1, characterized in that, The process of fusing the tissue density distribution map and the intrapulmonary CO2 concentration distribution map to generate a lung lobe-level ventilation thermogram includes: Spatial registration was performed between the tissue density distribution map and the lung CO2 concentration distribution map. The registered tissue density distribution map and lung CO2 concentration distribution map are fused at the pixel level. Tissue density is used as the structural layer and lung CO2 concentration is used as the functional layer. Through color mapping and overlay, a pseudo-color ventilation thermogram showing the ventilation status of different lung lobes is generated.

7. A dual-modality lung ventilation imaging system according to claim 6, characterized in that, The spatial registration of the tissue density distribution map and the lung CO2 concentration distribution map specifically includes: Initial rigid alignment: Initial rigid alignment is performed based on the physical coordinates of the detector array modules; Deformable registration: A deep learning registration network based on the U-Net architecture is used to predict the three-dimensional deformation vector field from the tissue density distribution map to the lung CO2 concentration distribution map, and a spatial transformation network is applied to the tissue density distribution map to achieve spatial alignment.

8. A dual-modal lung ventilation imaging system according to claim 7, characterized in that, The step of pixel-level fusion of the registered tissue density distribution map and the lung CO2 concentration distribution map includes: The registered tissue density distribution map and lung CO2 concentration distribution map are input into a preset deep learning fusion network, which includes a dual encoder, a fusion module and a single decoder connected in sequence. The tissue density distribution map and the lung CO2 concentration distribution map are respectively input into two independent branches of the dual encoder to extract the tissue density distribution feature map and the lung CO2 concentration feature map respectively; The fusion module is used to stitch the tissue density distribution feature map and the lung CO2 concentration distribution feature map in the channel dimension. The channel is recalibrated through the attention mechanism, and then the channel weights are output after passing through two fully connected layers. The single decoder uses skip connections to fuse tissue density distribution feature maps and lung CO2 concentration distribution feature maps, outputting a three-channel heat map.

9. A dual-modality lung ventilation imaging system according to claim 8, characterized in that, The loss function of the deep learning fusion network is: Where L1 represents pixel-level loss, Indicates perceived loss. Let λ1, λ2, and λ3 represent the structural similarity loss, and let λ1, λ2, and λ3 represent the weighting coefficients.

10. A dual-modal lung ventilation imaging method, characterized in that, Includes the following steps: S1. Control the first light source to emit light of the first wavelength to illuminate the target's chest cavity area; S2. Control the second light source to emit second-wavelength light to illuminate the target's chest cavity area; S3. Receive the light signals transmitted or reflected by the chest wall from the first light source module and the second light source module through the array detector module; wherein, the light signal corresponding to the first light source module includes the attenuation signal generated by the first band light in the lung tissue, and the light signal corresponding to the second light source module includes the CO2 absorption signal generated by the second band light in the lung. S4. The processor module reconstructs the lung tissue density distribution map based on the attenuation signal generated by the first band light in the lung tissue, and reconstructs the lung CO2 concentration distribution map based on the CO2 absorption signal generated by the second band light in the lung. S5. The processor module integrates the tissue density distribution map and the lung CO2 concentration distribution map to generate a ventilation thermogram at the lung lobe level.

Images