Multispectral fusion warm needle moxibustion applying depth and tissue response monitoring system

Abstract

The invention discloses a multispectral fusion warm needle moxibustion application depth and tissue response monitoring system, and belongs to the technical field of medical image processing and traditional Chinese medicine diagnosis and treatment assistance. The system comprises a multispectral image acquisition and adaptive registration module, a spectrum fusion and depth feature extraction module, an organization level response analysis and depth estimation module and a closed-loop feedback regulation and visual output module. Visible light, near infrared and thermal infrared spectrum images are synchronously collected, multi-scale spectrum fusion is carried out based on principal component analysis, multi-level thermal response accurate monitoring of subcutaneous tissue is achieved, tissue depth distribution is estimated through a deep learning network, a skin surface layer, a corium layer, a subcutaneous fat layer and a muscle fascia layer are divided, and the subcutaneous tissue is obtained. Thermal response parameters of each layer are calculated, a closed-loop feedback mechanism adaptively adjusts the collected parameters according to a monitoring result and optimizes moxibustion application suggestions, and standardized, precise and intelligent technical support is provided for needle warming moxibustion treatment.

Description

Technical Field

[0001] This invention belongs to the fields of medical image processing and TCM diagnosis and treatment assistance technology, specifically involving a multispectral fusion warm needle moxibustion depth and tissue response monitoring system, which is particularly suitable for non-invasive visualization monitoring and intelligent control of subcutaneous tissue heat conduction and physiological response during warm needle moxibustion treatment. Background Technology

[0002] Warm needling, an important component of traditional Chinese medicine, combines acupuncture with moxibustion. By igniting moxa wool on the needle handle while the needle is in place, heat is conducted through the needle to acupoints and deep tissues, achieving therapeutic effects such as warming the meridians, dispelling cold, and promoting blood circulation. With the deepening of modern research in traditional Chinese medicine, the mechanism of action of warm needling has gradually attracted attention. Studies have shown that the efficacy of warm needling is closely related to the depth of heat conduction in subcutaneous tissues, the temperature distribution pattern, and the thermal response characteristics of different tissue layers.

[0003] However, due to the lack of effective tissue depth monitoring technology, current warm acupuncture treatments mainly rely on the physician's clinical experience to control moxibustion parameters, making it difficult to accurately assess and real-time regulate the thermal penetration state of subcutaneous tissue. While traditional infrared thermal imaging technology can non-invasively acquire skin surface temperature distribution, it only reflects superficial thermal response information and cannot detect temperature changes and physiological responses in deeper tissues. This makes it difficult to quantify and assess the deep tissue effects of warm acupuncture treatment, affecting the personalized development of treatment plans and the objective evaluation of efficacy.

[0004] Chinese invention patent application CN119131120A discloses a method and system for measuring the area of ​​burn regions. This technology achieves area measurement of burn regions at different depths through registration of multimodal burn images, depth data extraction, and construction of three-dimensional surface models. While this technology has significant application value in burn medicine, it suffers from the following shortcomings: First, this technology focuses on static structural measurement of burn wounds, emphasizing three-dimensional morphological reconstruction and area calculation, without addressing dynamic monitoring of the temperature field or real-time analysis of heat conduction processes. Therefore, it is not suitable for monitoring tissue response during thermotherapy processes such as warm acupuncture. Second, the multimodal image fusion in this technology primarily serves three-dimensional geometric modeling, lacking a deep fusion strategy for multispectral information. It fails to fully utilize the differentiated characterization capabilities of different wavelength spectra for tissue layers, resulting in limited extraction efficiency of deep tissue information. Third, this technology employs a unidirectional image processing flow, lacking a feedback control mechanism based on monitoring results. It cannot adaptively optimize acquisition parameters or provide treatment parameter suggestions according to tissue response states, thus limiting the system's application potential in dynamic treatment scenarios.

[0005] Therefore, there is an urgent need to develop a warm needle acupuncture monitoring system that can integrate multispectral information, achieve precise monitoring of subcutaneous tissue depth response, and has closed-loop feedback regulation capabilities, in order to overcome the limitations of existing technologies and provide technical support for the standardization, precision, and intelligence of warm needle acupuncture treatment. Summary of the Invention

[0006] The purpose of this invention is to provide a multispectral fusion-based monitoring system for the depth and tissue response of warm needle moxibustion, aiming to solve the problems of insufficient acquisition of deep tissue information, low efficiency of multispectral information utilization, and lack of intelligent feedback and control mechanisms in existing warm needle moxibustion monitoring technologies, and to achieve accurate monitoring and intelligent optimization of subcutaneous tissue heat conduction and multi-level physiological responses during warm needle moxibustion.

[0007] This invention provides a multispectral fusion-based system for monitoring the depth and tissue response of warm needle acupuncture, including a multispectral image acquisition and adaptive registration module, a spectral fusion and depth feature extraction module, a tissue-level response analysis and depth estimation module, and a closed-loop feedback control and visualization output module.

[0008] The multispectral image acquisition and adaptive registration module simultaneously acquires visible light, near-infrared, and thermal infrared images of the area where the warm needle acupuncture is applied, forming a multispectral image set. Based on the feature point detection algorithm, the module performs spatial registration on each spectral image to eliminate spatial misalignment between different sensors, providing aligned multispectral data for subsequent fusion processing.

[0009] The spectral fusion and depth feature extraction module performs multi-scale spectral fusion on the registered multispectral image set based on the principal component analysis algorithm. It integrates complementary information from the visible, near-infrared and thermal infrared bands to generate a fused spectral image containing rich tissue layer features, and further extracts spectral-spatial depth features that characterize different tissue depths.

[0010] The tissue-level response analysis and depth estimation module uses a deep learning network to estimate the depth distribution of subcutaneous tissue from the depth feature map, and combines physiological prior knowledge to divide the tissue into the epidermis, dermis, subcutaneous fat layer and muscle fascia layer, calculates the thermal response parameters of each layer, and realizes the accurate quantification of the penetration of heat energy of warm needle moxibustion in different tissue layers.

[0011] The closed-loop feedback control and visualization output module calculates the tissue thermal energy penetration depth and response intensity evaluation index based on multi-level tissue response data, generates a three-dimensional visualization depth response map, and automatically generates collection parameter adjustment instructions and moxibustion parameter optimization suggestions by comparing the deviation of the evaluation index with the preset treatment target, forming a closed-loop control mechanism of monitoring-evaluation-feedback, realizing continuous optimization of system performance and intelligent assistance of treatment plan.

[0012] Through the above technical solutions, this invention achieves deep fusion of multispectral information, precise monitoring of multi-level subcutaneous tissue response, and intelligent feedback regulation based on monitoring results, providing innovative technical means for the scientific evaluation and precise implementation of warm acupuncture treatment.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] First, this invention simultaneously acquires images from three spectra: visible light, near-infrared, and thermal infrared, and performs multi-scale spectral fusion based on principal component analysis, fully utilizing the differentiated characterization capabilities of different wavelength bands for tissue layers. Visible light images provide information on surface morphology and blood vessel distribution, near-infrared images penetrate the skin surface to reveal subcutaneous blood flow and tissue structure, and thermal infrared images reflect temperature field distribution. The fusion of these three images significantly enhances the ability to extract information from deep tissues, improving tissue depth identification accuracy by more than 30% compared to single-spectrum monitoring.

[0015] Secondly, the tissue-level response analysis and depth estimation module designed in this invention uses a deep learning network to estimate tissue depth distribution from fused spectral features. Combined with prior physiological knowledge, it automatically stratifies the skin surface, dermis, subcutaneous fat layer, and muscle fascia layer, enabling quantitative calculation of thermal response parameters such as the rate of temperature change, cumulative heat energy, and response time constant for each layer. Clinical validation shows that this system can detect and visualize tissue thermal response changes up to 15 mm deep, providing unprecedented deep tissue observation capabilities for traditional Chinese medicine warm needling therapy. This significantly improves the precision of acupuncturists' control over the treatment process and their ability to evaluate treatment effects.

[0016] Third, the closed-loop feedback control mechanism constructed in this invention automatically generates adjustment instructions for acquisition parameters and optimization suggestions for moxibustion parameters based on the deviation between multi-level tissue response data and preset treatment targets. This mechanism enables the system to adaptively optimize image acquisition strategies based on real-time monitoring results, improving its ability to capture dynamically changing tissue responses; simultaneously, it provides physicians with suggestions on moxibustion temperature, time, and depth based on objective data, realizing intelligent auxiliary decision-making for warm acupuncture treatment. Compared to the traditional one-way monitoring mode, the closed-loop feedback mechanism continuously improves monitoring accuracy during treatment, increasing the overall monitoring accuracy by more than 25%, providing technical support for the standardization and personalization of warm acupuncture treatment. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall architecture of the multispectral fusion-based warm needle moxibustion depth and tissue response monitoring system of the present invention.

[0018] Figure 2 This is a schematic diagram of the multispectral image acquisition and adaptive registration module of the present invention;

[0019] Figure 3 This is a schematic diagram of the processing flow of the spectral fusion and depth feature extraction module of the present invention;

[0020] Figure 4 This is a schematic diagram illustrating the working principle of the organization hierarchy response analysis and depth estimation module of the present invention;

[0021] Figure 5 This is a schematic diagram of the feedback control process of the closed-loop feedback control and visualization output module of the present invention;

[0022] Figure 6 This is a schematic diagram of the multi-scale pyramid structure for the multi-scale spectral fusion processing of the present invention. Detailed Implementation

[0023] Please refer to the attached document. Figures 1-6 To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] Reference Figure 1 This invention provides a multispectral fusion-based system for monitoring the depth of warm needle acupuncture and tissue response. The system includes a multispectral image acquisition and adaptive registration module 1, a spectral fusion and depth feature extraction module 2, a tissue level response analysis and depth estimation module 3, and a closed-loop feedback control and visualization output module 4.

[0025] The multispectral image acquisition and adaptive registration module 1 is used to simultaneously acquire visible light, near-infrared, and thermal infrared images of the area treated with warm needle acupuncture, obtaining a multispectral image set. Based on a feature point detection algorithm, it performs spatial registration on each spectral image in the multispectral image set, resulting in a registered multispectral image set. This module is the data source for the entire system, and its acquisition quality and registration accuracy directly affect the effectiveness of subsequent processing.

[0026] The spectral fusion and depth feature extraction module 2 is connected to the multispectral image acquisition and adaptive registration module 1. It performs multi-scale spectral fusion on the registered multispectral image set based on principal component analysis (PCA) to obtain a fused spectral image. Then, it extracts spectral-spatial depth features representing different tissue layers from the fused spectral image to obtain a depth feature map. This module achieves deep integration and feature extraction of multispectral information, providing high-quality feature representations for tissue depth estimation.

[0027] The tissue-level response analysis and depth estimation module 3 is connected to the spectral fusion and depth feature extraction module 2. It inputs the depth feature map into a pre-trained depth estimation network to obtain a tissue depth distribution map. Based on the tissue depth distribution map and prior physiological knowledge, it performs hierarchical segmentation of subcutaneous tissue to obtain multi-level tissue response data. This module is the core analysis unit of the system, realizing the intelligent conversion from spectral features to tissue depth information.

[0028] The closed-loop feedback control and visualization output module 4 is connected to the tissue-level response analysis and depth estimation module 3. It is used to calculate evaluation indicators of tissue thermal energy penetration depth and response intensity based on multi-level tissue response data, generate a three-dimensional visualized depth response map, and generate acquisition parameter adjustment instructions and moxibustion parameter optimization suggestions based on the deviation between the evaluation indicators and the preset treatment target. The acquisition parameter adjustment instructions are then fed back to the multispectral image acquisition and adaptive registration module 1, forming a closed-loop feedback control mechanism. This module not only realizes the visualization of monitoring results but, more importantly, constructs an intelligent feedback control mechanism based on the monitoring results.

[0029] Reference Figure 2 The multispectral image acquisition and adaptive registration module 1 includes a multispectral camera unit, a time synchronization unit, a feature point detection unit, and a registration calculation unit.

[0030] A multispectral camera unit is used to simultaneously acquire visible light, near-infrared, and thermal infrared images of the area treated with warm needle acupuncture. In a preferred embodiment of the invention, the visible light image wavelength range is 400nm to 700nm, acquired using an RGB camera with a resolution of 1920×1080 pixels, which can clearly present the skin surface morphology, blood vessel distribution, and acupuncture point location; the near-infrared image wavelength range is 700nm to 1400nm, acquired using an InGaAs near-infrared camera, which can penetrate the skin to a depth of 3mm to 5mm, reflecting subcutaneous hemodynamic changes and superficial tissue structures; the thermal infrared image wavelength range is 8μm to 14μm, acquired using an uncooled vanadium oxide microbolometer array sensor, with a temperature resolution of 0.05℃ and a spatial resolution of 640×480 pixels, which can accurately measure the skin surface temperature distribution. The cameras of the three spectral channels are arranged at a 120° angle, 30cm to 50cm away from the acupuncture area, with a field of view covering a monitoring area of ​​15cm×15cm.

[0031] The time synchronization unit is connected to the multispectral camera unit to precisely synchronize the image acquisition time of each spectral channel. Because the temperature field dynamically changes over time during warm needle acupuncture, time misalignment between different spectral images can lead to artifacts in the fused image. This invention employs a hardware-triggered synchronization method, sending trigger signals simultaneously to the three cameras via an FPGA (Field-Programmable Gate Array) controller to ensure that the difference in image acquisition time between the three spectral channels is less than 1ms. The trigger frequency is set to 5Hz to 10Hz, which captures the dynamic changes in the temperature field while avoiding data redundancy. Furthermore, the timestamp of each image is recorded by a high-precision crystal oscillator clock, achieving a time accuracy of 1μs, providing a reliable time reference for subsequent time-series analysis.

[0032] The feature point detection unit is connected to the time synchronization unit and is used to detect key feature points in various spectral images based on the scale-invariant feature transform (SIFT) algorithm. Traditional feature point detection algorithms are prone to failure due to significant differences in contrast, texture, and edge features among different spectral images. This invention employs an improved SIFT algorithm. First, adaptive histogram equalization is performed on each spectral image to enhance image contrast. Then, local extrema are detected in scale space as candidate feature points. For visible light images, high-contrast features around vascular intersections, skin texture corners, and needle pricks are primarily detected. For near-infrared images, the focus is on detecting subcutaneous vascular nodes and tissue structure boundaries. For thermal infrared images, temperature gradient peaks and isotherm inflection points are extracted as features. 200 to 500 feature points are extracted from each image to ensure sufficient matching point pairs during image registration.

[0033] The registration calculation unit is connected to the feature point detection unit and is used to calculate the spatial transformation matrix between different spectral images based on key feature points to complete image registration. This invention employs a robust registration strategy based on the RANSAC (Random Sample Consensus) algorithm. First, feature matching is performed using feature point descriptors (128-dimensional SIFT descriptors), and reliable matching point pairs are screened using the nearest neighbor distance ratio threshold method, with the threshold set to 0.7 to 0.8. Then, the RANSAC algorithm is used to estimate the affine transformation matrix from the source image to the target image, which includes four transformation parameters: translation, rotation, scaling, and shearing. The RANSAC iteration count is set to 1000 times, and the inlier threshold is set to 2 pixels, effectively eliminating mismatched point pairs and improving registration accuracy. Finally, bilinear interpolation is used to resample the source image and transform it to the coordinate system of the target image to complete spatial registration. In an embodiment of the present invention, a thermal infrared image is used as a reference image for registration. Both the visible light image and the near-infrared image are registered to the coordinate system of the thermal infrared image. The alignment error of the registered image is less than 1 pixel, which provides a precise spatial alignment basis for subsequent pixel-level spectral fusion.

[0034] The registered multispectral image set contains aligned images of three channels, with each pixel location corresponding to the response characteristics of the same spatial point in the moxibustion area under different spectral bands. This multispectral image set is not only spatially aligned but also temporally synchronized, providing high-quality input data for the spectral fusion and deep feature extraction module 2.

[0035] Furthermore, to adapt to different lighting conditions and patient skin color variations, the multispectral image acquisition and adaptive registration module 1 also incorporates an adaptive exposure control mechanism. This mechanism automatically adjusts the exposure time and gain parameters of each camera based on the brightness histogram distribution of the current image, ensuring that the image brightness distribution remains within the middle of the dynamic range and avoiding information loss due to overexposure or underexposure. The exposure time range for the visible light camera is 1ms to 20ms, for the near-infrared camera it is 5ms to 50ms, and the thermal infrared camera employs automatic gain control, with a temperature measurement range covering 20℃ to 60℃.

[0036] Reference Figure 3 The spectral fusion and deep feature extraction module 2 includes a principal component analysis unit, a multi-scale fusion unit, and a deep feature encoding unit.

[0037] The principal component analysis (PCA) unit is used to perform PCA on the registered multispectral image set to extract principal component features. PCA is a classic data dimensionality reduction and feature extraction method that can convert multiple correlated variables into a few uncorrelated principal components while retaining the main information in the original data. In this invention, the registered image data of the visible light, near-infrared, and thermal infrared spectral channels are organized into a high-dimensional data matrix, where each pixel position forms a three-dimensional spectral vector. PCA analysis is performed on the spectral vectors of all pixels to calculate the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvalues ​​are sorted from largest to smallest, and the top k principal components are selected as the spectral feature representations. In a preferred embodiment, the number of principal components k is set to 3 to 8, adaptively determined according to the cumulative variance contribution rate, requiring the cumulative variance contribution rate to be no less than 95%. This preserves the main information in the original spectral data, achieves data dimensionality reduction, and reduces the computational complexity of subsequent processing.

[0038] The physical meaning of the principal components is closely related to the spectral response characteristics. The first principal component typically corresponds to the overall intensity of the temperature field, reflecting the average temperature level of the moxibustion area; the second principal component corresponds to the contrast between visible and near-infrared light, highlighting the distribution of subcutaneous blood vessels and hemodynamic changes; the third principal component corresponds to the spatial gradient of the temperature field, amplifying the differences in thermal response between different tissue layers. Through principal component analysis, redundant information in the original multispectral image is removed, while feature information sensitive to tissue depth response is enhanced, laying a feature foundation for subsequent fusion processing and depth estimation.

[0039] The multi-scale fusion unit is connected to the principal component analysis unit to construct a multi-scale Gaussian pyramid, performing weighted fusion of principal component features at different scales. (See reference...) Figure 6 The multi-scale fusion processing employs an image pyramid architecture. First, Gaussian filtering and downsampling are applied to the principal component images to construct an L-layer Gaussian pyramid (L is typically 4 to 6), with each layer having half the resolution of the previous layer. At each layer of the pyramid, different principal component images are fused using an adaptive weighting strategy.

[0040] The fusion weights are determined based on the saliency measure of each principal component image at the current scale. This invention uses local energy and local entropy as saliency indicators. Local energy is measured by calculating the sum of squared pixel values ​​of image patches, reflecting the contrast of image regions; local entropy is measured by calculating the gray-level histogram entropy of image patches, reflecting the information richness of image regions. For the principal component images of the l-th layer of the pyramid... Its position Fusion weights at the point The calculation is as follows:

[0041] ,

[0042] in, The position of the i-th principal component in the l-th layer of the pyramid. Normalized local energy at the point, To normalize the local entropy, The energy and entropy balance coefficient is preferably set to 0.6, which prioritizes regions with high contrast. Both local energy and local entropy are within the range... The calculation is performed within the window, and normalization is achieved by dividing by the sum of the corresponding indices of all principal components, ensuring that the sum of the weights of all principal components at the same position is 1.

[0043] fused images The result is obtained by weighted summation of the principal component images:

[0044] ,

[0045] Here, k represents the number of principal components. This formula achieves adaptive fusion of principal component features at each level of the pyramid, with principal components that have high contrast and rich information having a greater weight in the fusion result.

[0046] The fused images of each layer of the multi-scale pyramid represent the tissue response characteristics at different spatial frequencies. The top layer (low-resolution layer) reflects the macroscopic distribution of the temperature field and large-scale tissue structure, while the bottom layer (high-resolution layer) preserves the fine variations in temperature gradient and small-scale vascular features. To generate a fused spectral image containing full-band information, inverse reconstruction of each layer of the pyramid is required. This invention employs the Laplacian pyramid reconstruction method, starting from the top layer and upsampling and fusing layer by layer until the original resolution is restored, resulting in the final fused spectral image. Each layer of the Laplacian pyramid represents the difference between the current layer and the next layer after upsampling, and this difference contains detailed information at that scale. The inverse reconstruction process accumulates the detailed information from each scale into the fused result, achieving complete preservation of multi-scale information.

[0047] The deep feature encoding unit is connected to the multi-scale fusion unit to extract spectral-spatial depth features of the fused spectral image based on a convolutional neural network. This invention uses ResNet-50 as the backbone network for feature extraction. This network contains 49 convolutional layers and 1 fully connected layer, and solves the gradient vanishing problem in deep networks through a residual connection mechanism, enabling it to learn rich image semantic features.

[0048] To adapt to the specific task of monitoring warm needle acupuncture, this invention improves ResNet-50. First, in the network's input layer, the fused spectral image is used as a single-channel input, and then... Convolutional and max-pooling layers are used for initial feature extraction. Then, features are extracted layer by layer through four groups of residual blocks (each group containing multiple residual units), progressively reducing the spatial resolution of the feature maps to a fraction of the original image's resolution. , , and The number of feature channels increases sequentially to 64, 128, 256, and 512. The design of residual block groups enables the network to extract multi-level representations from low-level texture features to high-level semantic features at different depths.

[0049] Following the third and fourth residual block groups of ResNet-50, this invention introduces spatial attention and channel attention mechanisms to enhance the network's focus on tissue-depth-related features. The spatial attention mechanism generates a spatial attention weight map by performing global average pooling and max pooling on the feature map, highlighting regions with significant temperature gradients. The channel attention mechanism generates a channel attention weight vector through adaptive average pooling and fully connected layers, strengthening feature channels sensitive to tissue layers. The two attention weights are multiplied by the original feature map to obtain an attention-enhanced feature representation.

[0050] The output of ResNet-50 is a 512-channel depth feature map with a spatial resolution equal to that of the original image. This feature map encodes high-level semantic information related to tissue depth in the fused spectral image, including temperature response patterns, thermal penetration boundaries, and hemodynamic features at different tissue levels. As input to the tissue-level response analysis and depth estimation module 3, the depth feature map provides rich feature representations for accurate depth estimation.

[0051] It should be noted that ResNet-50 is trained using a transfer learning strategy in this invention. First, it is pre-trained on the large-scale ImageNet image dataset to learn general image feature representation capabilities. Then, it is fine-tuned on a multispectral image dataset of warm needle acupuncture treatment to adapt the network to specific spectral features and tissue response patterns. The training dataset contains 1000 multispectral image sequences of warm needle acupuncture treatment processes, covering different patients, different acupoints, and different acupuncture parameters. Each sample is labeled with the corresponding ground truth value of tissue depth (obtained through high-frequency ultrasound measurement). The training process uses the Adam optimizer with a learning rate of 0.0001, a batch size of 16, and is trained for 100 epochs until the loss function converges.

[0052] Reference Figure 4 The organization-level response analysis and depth estimation module 3 includes a depth estimation network unit, an organization-level unit, and a thermal response calculation unit.

[0053] Depth estimation network units are used in deep learning networks based on encoder-decoder architectures to estimate the depth distribution of tissues from depth feature maps. This invention adopts the U-Net architecture as the basic framework of the depth estimation network. This architecture consists of a symmetrical encoder and decoder, and features are transferred between corresponding layers of the encoder and decoder through skip connections, effectively fusing low-level detail features and high-level semantic features, making it suitable for dense prediction tasks.

[0054] The encoder portion of U-Net consists of 5 downsampling layers, each containing two layers. Convolutional layer and one Max pooling layers are used as input to the encoder, which takes a 512-channel depth feature map output from the deep feature encoding unit. Through progressive downsampling, the spatial resolution of the feature map decreases sequentially, while the number of channels increases sequentially to 512, 512, 1024, 1024, and 2048, extracting multi-scale feature representations from local texture to global semantics. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function to improve the network's training stability and convergence speed.

[0055] The U-Net decoder consists of 5 upsampling layers, each containing one Transposed convolutional layer (for upsampling) and two Convolutional layers. The decoder recovers the spatial resolution of the feature maps through progressive upsampling, and at each level, it concatenates the feature maps with those of the corresponding encoder layers, fusing high-resolution detail features and low-resolution semantic features. The concatenated feature maps are then further processed by convolutional layers to extract fused features, with the number of feature channels decreasing sequentially to 1024, 512, 256, 128, and 64. Finally, in the last stage... The convolutional layer outputs a single-channel tissue depth distribution map with the same spatial resolution as the input depth feature map.

[0056] Each pixel value in the tissue depth distribution map represents the thermal response depth of the tissue at that location, in millimeters (mm). The depth values ​​range from 0 mm to 20 mm, where 0 mm corresponds to the skin surface and 20 mm corresponds to the maximum detection depth. The depth estimation network is trained through end-to-end supervised learning, and the loss function is a weighted combination of mean squared error loss and edge preservation loss.

[0057] ,

[0058] in, To predict the mean squared error between the depth map and the true depth map. Edge-preserving loss is used to penalize prediction errors in regions of depth discontinuity. The weighting coefficient for the edge loss is preferably 0.1. The mean squared error loss is defined as:

[0059] ,

[0060] Where N is the total number of pixels. Let i be the true depth value of the i-th pixel. To predict depth values, the edge-preserving loss computes the gradient of the depth map based on the Sobel operator and measures the difference between the predicted depth gradient and the true depth gradient.

[0061] ,

[0062] in, Represents the gradient operator. Let L1 norm be denoted. By jointly optimizing the mean squared error and edge preservation loss, the depth estimation network can not only accurately predict depth values, but also maintain the clarity of tissue hierarchy boundaries and avoid over-smoothing of the depth map.

[0063] In practical applications, the depth estimation network is trained offline on the training dataset, and the network parameters are fixed after training for online inference. Given an input depth feature map, the depth estimation network outputs a tissue depth distribution map within approximately 50ms, meeting the requirements of real-time monitoring.

[0064] The tissue stratification unit is connected to the depth estimation network unit to divide the subcutaneous tissue into the epidermis, dermis, subcutaneous fat layer, and muscle fascia layer based on tissue depth distribution maps and prior physiological knowledge. Tissue stratification is a depth-threshold-based hierarchical division process that incorporates anatomical knowledge of skin physiology.

[0065] According to histological studies of skin, human skin is divided into three main layers from the outside in: the epidermis, the dermis, and the subcutaneous tissue. The epidermis is approximately 0.05 mm to 0.2 mm thick, the dermis is approximately 1 mm to 4 mm thick, and the thickness of the subcutaneous tissue (including the fat layer and fascia layer) varies depending on the location, generally ranging from 2 mm to 15 mm. During warm needle moxibustion, the heat energy first acts on the surface layer of the skin (including the epidermis and superficial dermis), and then conducts to the deeper layers, successively affecting the dermis, subcutaneous fat layer, and muscle fascia layer.

[0066] This invention uses the following depth thresholds for tissue stratification based on tissue depth distribution maps: the skin surface layer has a depth range of 0mm to 2mm, mainly including the epidermis and superficial dermis. This area responds rapidly to thermal stimulation and exhibits significant temperature changes; the dermis layer has a depth range of 2mm to 5mm, including the deep dermis and subcutaneous transition zone. This area is rich in blood vessels, and the thermal response manifests as hemodynamic changes; the subcutaneous fat layer has a depth range of 5mm to 10mm. Adipose tissue has low thermal conductivity, and heat conduction is slow in this area, resulting in a delayed temperature rise; the muscle fascia layer has a depth range of 10mm to 15mm. This area is the target treatment depth for warm needle moxibustion. Muscle and fascia tissues are sensitive to temperature, and the thermal response is closely related to the treatment effect.

[0067] The tissue stratification process determines the tissue layer to which each pixel belongs based on its depth value, generating a four-layer tissue mask map. Each mask map identifies the spatial distribution area of ​​the tissue in that layer. Through tissue stratification, the originally continuous depth distribution information is transformed into a physiologically meaningful hierarchical structure, providing an anatomical basis for subsequent thermal response parameter calculations and treatment assessments.

[0068] The thermal response calculation unit is connected to the tissue layering unit and is used to calculate the temperature change rate, heat accumulation, and response time constant of each tissue layer. Thermal response parameters are key indicators for quantitatively evaluating the effect of warm needling moxibustion, reflecting the response characteristics of different tissue layers to thermal stimulation.

[0069] The rate of temperature change characterizes the speed at which tissue temperature changes over time, and is defined as the increment of temperature per unit time. For the l-th layer of tissue, the rate of temperature change at time t is... The calculation is as follows:

[0070] ,

[0071] in, The average temperature of the l-th layer of tissue at time t is obtained by averaging the pixel values ​​within the masked region of the thermal infrared image. The sampling time interval is set to 0.2s to 1.0s in this invention. The unit of temperature change rate is ℃ / s, with positive values ​​indicating heating and negative values ​​indicating cooling. By monitoring the temperature change rate, the current stage of moxibustion can be determined. For example, the temperature change rate is larger in the early stage of moxibustion, while the temperature tends to stabilize and the change rate decreases after continuous moxibustion.

[0072] The cumulative heat energy represents the total heat absorbed by the tissue during moxibustion and is an important indicator for assessing the degree of heat penetration. For the first layer of tissue, starting from the beginning of moxibustion (time... The cumulative amount of heat energy up to the current time t The calculation is as follows:

[0073] ,

[0074] in, The specific heat capacity of the l-th layer is... The mass of this layer of tissue is approximately estimated from its volume and density. For a moment The rate of temperature change. In numerical calculations, the integral is approximated using discrete summation:

[0075] ,

[0076] Where n represents the number of samples taken from the start of moxibustion to the current moment. The time of the i-th sampling is given. The unit of accumulated heat energy is joules (J). By comparing the accumulated heat energy at different tissue layers, the depth of heat penetration in the tissue can be assessed; a larger accumulated amount indicates a stronger thermal stimulus to that tissue layer.

[0077] The response time constant characterizes the rate of tissue response to thermal stimuli and is defined as the time required for the tissue temperature to rise from its initial value to 63.2% of its steady-state value. The response time constant reflects the thermal inertia of the tissue; a smaller time constant indicates a faster tissue response. For the l-th layer of tissue, it is assumed that its temperature response follows a first-order exponential model:

[0078] ,

[0079] in, This refers to the initial temperature of the tissue layer before moxibustion. This refers to the steady-state temperature of the tissue layer after moxibustion. The response time constant is used. Parameters are estimated by nonlinear fitting of the measured temperature curve. and The unit of response time constant is seconds (s). Due to the high thermal conductivity and low heat capacity of the skin surface, the response time constant is usually 10s to 30s; while due to the long heat conduction path and high heat capacity of deeper tissues, the response time constant can reach 60s to 120s.

[0080] By calculating the three thermal response parameters mentioned above, the system can comprehensively quantify the thermal response characteristics of each tissue level, generating multi-level tissue response data. This data not only provides physicians with objective criteria for evaluating treatment effectiveness but also supports intelligent decision-making in the closed-loop feedback control and visualization output module 4.

[0081] Reference Figure 5 The closed-loop feedback control and visualization output module 4 includes an evaluation index calculation unit, a three-dimensional visualization unit, a deviation analysis unit, and a parameter optimization unit.

[0082] The evaluation index calculation unit is used to calculate tissue thermal energy penetration depth and response intensity evaluation indices based on multi-level tissue response data. Tissue thermal energy penetration depth refers to the maximum tissue depth to which thermal energy can effectively act, defined as the maximum depth to which the temperature rise exceeds the baseline temperature by 2°C. Response intensity evaluation indices include peak temperature gradient, thermal energy accumulation rate, and tissue response uniformity.

[0083] The peak temperature gradient characterizes the steepness of the temperature distribution within the tissue, and is obtained by calculating the maximum rate of temperature change along the depth direction:

[0084] ,

[0085] in, Let be the temperature at depth d. This represents the temperature gradient along the depth direction. The unit for the peak temperature gradient is °C / mm. A large peak temperature gradient indicates a significant temperature difference between different tissue layers, suggesting that the penetration of heat energy at a certain depth is hindered, and it may be necessary to adjust the moxibustion parameters to improve the heat transfer of deep tissues.

[0086] The rate of heat accumulation characterizes the speed at which tissue absorbs heat energy per unit time, and is defined as the derivative of the accumulated heat energy with respect to time.

[0087] ,

[0088] In numerical calculations, the difference approximation is used:

[0089] ,

[0090] in, This represents the accumulated heat energy of the l-th layer of tissue at time t. The unit of heat energy accumulation rate is watts (W). Monitoring the heat energy accumulation rate can assess the heat conduction efficiency of moxibustion in real time. If the rate is too low, it indicates that the intensity of moxibustion needs to be increased or the moxibustion time needs to be extended.

[0091] Tissue response uniformity characterizes the degree of balance in thermal response among different tissue layers, and is obtained by calculating the coefficient of variation of the temperature change rate of each layer:

[0092] ,

[0093] in, The standard deviation of the temperature change rate for each layer. This represents the average rate of temperature change for each layer. Tissue response uniformity is a dimensionless index; the smaller the value, the more uniform the thermal response of each tissue layer, and the more ideal the treatment effect.

[0094] The 3D visualization unit is connected to the evaluation index calculation unit to generate a 3D visualized depth response map. 3D visualization combines tissue depth information with thermal response parameters to present the thermal energy distribution and response characteristics of subcutaneous tissue during warm needle moxibustion in an intuitive 3D graphical format.

[0095] 3D visualization employs volume rendering technology to map tissue depth and temperature distribution maps into three-dimensional space. Specifically, using the skin surface of the moxibustion area as a reference plane, tissue layers are superimposed along the depth direction to form three-dimensional voxel data. The position of each voxel corresponds to the spatial coordinates of the tissue, and the color of the voxel encodes the temperature value, using a color mapping from blue (low temperature) to red (high temperature). By adjusting the viewing angle and transparency, physicians can observe the temperature distribution at different depths and identify the pathways and bottleneck areas of heat penetration.

[0096] In addition, temperature isosurfaces and heat flow lines are overlaid on the 3D visualization. Temperature isosurfaces connect spatial points with the same temperature value to form curved surfaces, which intuitively show the geometry of the temperature field; heat flow lines are drawn along the direction of heat conduction, with arrows pointing in the direction of heat flow, and the line thickness representing heat flux density, revealing the conduction path and distribution pattern of heat energy in the tissue.

[0097] The 3D visualization depth response map is presented in an interactive interface, allowing physicians to rotate, zoom, and section the image using a mouse or touchscreen to observe tissue response details from different angles. The visualization interface also updates in real time; as the moxibustion process progresses, the 3D map dynamically reflects the evolution of the temperature field, providing physicians with a comprehensive, multi-angle monitoring perspective.

[0098] The deviation analysis unit is connected to the evaluation index calculation unit and is used to calculate the deviation between the evaluation index and the preset treatment target. The preset treatment target is set in advance according to the patient's condition, physical condition and acupoint characteristics, including the target thermal energy penetration depth, the target temperature gradient peak value and the target tissue response uniformity, etc.

[0099] The formula for calculating deviation is:

[0100] ,

[0101] in, Let be the measured value of the i-th evaluation index. To preset the target value, This represents the percentage of relative deviation. A positive deviation indicates that the measured value is higher than the target value, while a negative deviation indicates that the measured value is lower than the target value.

[0102] The results of deviation analysis are used to determine whether the current moxibustion effect has met expectations and in which aspects adjustments are needed. For example, if the deviation of the heat penetration depth is negative and the absolute value is large, it indicates that the heat energy has not effectively reached the target depth, and the moxibustion intensity or duration needs to be increased. If the deviation of the temperature gradient peak value is positive and the absolute value is large, it indicates that the temperature difference between tissue layers is too large, which may lead to overheating of superficial tissues and insufficient heating of deep tissues, and the moxibustion strategy needs to be optimized to improve the uniformity of heat energy distribution.

[0103] The parameter optimization unit is connected to the deviation analysis unit and is used to generate adjustment instructions for the acquired parameters and optimization suggestions for the moxibustion parameters based on the deviation. The parameter optimization adopts a strategy that combines rule-based expert systems and data-driven machine learning methods.

[0104] The acquisition parameter adjustment commands target the acquisition strategy of the multispectral image acquisition and adaptive registration module 1, including exposure time adjustment, gain adjustment, and sampling frequency adjustment. When deviation analysis shows a large temperature change rate, the sampling frequency is increased to capture dynamic changes more densely; when the image brightness is insufficient or overexposed, the exposure time and gain are adjusted to optimize image quality. The adjustment commands are sent to module 1 through the feedback channel to achieve adaptive optimization of the acquisition parameters.

[0105] The moxibustion parameter optimization suggestions are tailored to the physician's moxibustion procedure, including recommendations for moxibustion temperature, moxibustion time, and acupuncture depth. These suggestions are generated based on bias analysis and an expert knowledge base. For example, if the heat penetration depth is insufficient, it is recommended to appropriately increase the amount of moxa stick used or shorten the distance between the moxa stick and the skin to increase the moxibustion temperature, or to extend the single moxibustion time to increase heat accumulation; if the thermal response of a certain tissue layer is insufficient, it is recommended to adjust the acupuncture depth so that the needle tip is closer to that tissue layer to enhance the direct conduction of heat energy.

[0106] The parameter optimization unit also integrates an intelligent optimization algorithm based on reinforcement learning. During treatment, the system continuously learns the mapping relationship between deviations and parameter adjustments, gradually optimizing the decision-making strategy through a trial-and-error mechanism, making parameter adjustments more precise and effective. The reinforcement learning model adopts a deep Q-network (DQN) architecture, using evaluation indicators and deviations as state inputs, parameter adjustments as action outputs, and the degree of improvement in treatment effect as a reward signal. Through online learning with multiple patients, the system's parameter optimization capabilities have continuously improved, providing intelligent support for achieving personalized precision treatment.

[0107] Through the collaborative work of the four units mentioned above, the closed-loop feedback control and visualization output module 4 achieves a complete closed loop from monitoring data to visualization presentation, and from deviation analysis to parameter optimization. This closed-loop mechanism not only improves the system's monitoring accuracy and robustness, but also provides objective data support and intelligent suggestions for physicians' clinical decisions, promoting the standardization, precision, and personalization of warm acupuncture treatment.

[0108] The overall workflow of the multispectral fusion-based warm needling moxibustion depth and tissue response monitoring system of the present invention is as follows:

[0109] Before the warm needle moxibustion treatment begins, the physician sets the treatment target parameters, including the target heat penetration depth, the target temperature range, and the target moxibustion time. The system initializes the parameters of each module according to the treatment target.

[0110] After treatment begins, the multispectral image acquisition and adaptive registration module 1 simultaneously acquires visible light, near-infrared, and thermal infrared images at a frequency of 5Hz to 10Hz, completes image registration, and generates a registered multispectral image set. This process takes approximately 100ms to 200ms.

[0111] The registered multispectral image set is transmitted to the spectral fusion and depth feature extraction module 2, which performs principal component analysis, multi-scale spectral fusion, and depth feature extraction to generate a depth feature map. This process takes approximately 150ms to 300ms.

[0112] The depth feature map is input to the tissue-level response analysis and depth estimation module 3. This module outputs a tissue depth distribution map through a depth estimation network, and performs tissue stratification and thermal response parameter calculations to generate multi-level tissue response data. This process takes approximately 50ms to 100ms.

[0113] Multi-level tissue response data is transmitted to the closed-loop feedback control and visualization output module 4. This module calculates evaluation indicators, generates a three-dimensional visualization depth response map, performs deviation analysis, and generates acquisition parameter adjustment instructions and moxibustion parameter optimization suggestions. The acquisition parameter adjustment instructions are sent to module 1 through the feedback channel to adjust the image acquisition strategy for the next cycle; the moxibustion parameter optimization suggestions are presented to the physician on the interface to assist the physician in optimizing the moxibustion operation. This process takes approximately 100ms to 200ms.

[0114] The entire system has a processing cycle of approximately 400ms to 800ms, meeting the requirements for real-time monitoring. As treatment progresses, the system continuously executes the above process in a loop, dynamically monitoring tissue response, updating the visualization in real time, and adaptively optimizing parameters based on deviations, forming a closed-loop control mechanism of monitoring-evaluation-feedback.

[0115] After treatment, the system generates a complete treatment report, including temperature curves, heat energy accumulation curves, evaluation index change curves, and keyframes of a three-dimensional visualization depth response map for each tissue layer during the moxibustion process, providing data support for physicians to evaluate treatment effectiveness and formulate subsequent treatment plans.

[0116] To verify the effectiveness of the system of this invention, a clinical application test was conducted in the acupuncture department of a traditional Chinese medicine hospital. Thirty patients with chronic low back pain were selected and treated with warm acupuncture at the Shenshu (BL23) and Dachangshu (BL25) acupoints in the lower back. The treatment process was monitored in real time using the system of this invention, and a high-frequency ultrasound device (probe frequency 18MHz, detection depth 20mm) was used as a reference standard to verify the accuracy of the tissue depth distribution and thermal response parameters estimated by the system.

[0117] Experimental results show that the average absolute error between the tissue depth distribution estimated by the system of this invention and the ultrasound measurement results is 0.8 mm, and the relative error is less than 10%, indicating high accuracy in depth estimation. The four tissue layers (epidermal layer, dermis, subcutaneous fat layer, and muscle fascia layer) identified by the system correspond well with the anatomical layers in the ultrasound images, and the tissue layering accuracy reaches 92%.

[0118] During the warm needle moxibustion treatment, the system monitored the dynamic process of heat energy conduction from the skin surface to deeper layers in real time. Within 30 seconds of the start of moxibustion, the skin surface temperature rapidly rose to 40℃ to 45℃, with a temperature change rate of 0.5℃ / s to 0.8℃ / s. As the moxibustion continued, the heat energy gradually penetrated into the dermis and subcutaneous fat layer. At 2 minutes, the dermal temperature rose to 38℃ to 42℃, and the subcutaneous fat layer temperature rose to 36℃ to 39℃. At 5 minutes, the heat energy reached the muscle fascia layer, where the temperature rose to 35℃ to 37℃, achieving a tissue thermal response to a depth of 12mm to 15mm. The average heat penetration depth calculated by the system was 13.5mm, consistent with the preset treatment target (12mm to 15mm), indicating that the moxibustion effect met expectations.

[0119] Regarding closed-loop feedback control, the system automatically adjusts the acquisition parameters based on real-time monitoring deviations. During the treatment of a certain patient, local overexposure of the visible light image was caused by skin reflection. The system automatically shortened the exposure time from 10ms to 6ms and reduced the gain to improve image quality. Simultaneously, the system detected that the patient had a thick subcutaneous fat layer, resulting in slower heat penetration. It recommended that the physician appropriately extend the moxibustion time. After the physician adopted the suggestion, the final heat penetration depth reached the treatment target.

[0120] Clinical trials have shown that the system of this invention can accurately monitor the depth response of subcutaneous tissue during warm needle moxibustion, provide real-time visual feedback and intelligent suggestions, significantly improve the accuracy and objectivity of treatment, and has been highly recognized by clinicians.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-spectral fusion system for depth and tissue response monitoring of warm needling, characterized in that, include: The multispectral image acquisition and adaptive registration module is used to simultaneously acquire visible light images, near-infrared images, and thermal infrared images of the area where the warm needle acupuncture is applied, to obtain a multispectral image set, and to perform spatial registration of each spectral image in the multispectral image set based on a feature point detection algorithm, to obtain a registered multispectral image set. The spectral fusion and depth feature extraction module is connected to the multispectral image acquisition and adaptive registration module. It is used to perform multi-scale spectral fusion on the registered multispectral image set based on the principal component analysis algorithm to obtain a fused spectral image, and extract spectral-spatial depth features representing different tissue levels from the fused spectral image to obtain a depth feature map. The tissue-level response analysis and depth estimation module is connected to the spectral fusion and depth feature extraction module. It is used to input the depth feature map into a pre-trained depth estimation network to obtain a tissue depth distribution map. Based on the tissue depth distribution map and physiological prior knowledge, the subcutaneous tissue is divided into layers to obtain multi-level tissue response data. The multi-level tissue response data includes thermal response parameters of the skin surface, dermis, subcutaneous fat layer and muscle fascia layer. The closed-loop feedback control and visualization output module is connected to the tissue-level response analysis and depth estimation module. It is used to calculate the tissue thermal energy penetration depth and response intensity evaluation index based on the multi-level tissue response data, generate a three-dimensional visualization depth response map, and generate acquisition parameter adjustment instructions and moxibustion parameter optimization suggestions according to the deviation between the evaluation index and the preset treatment target. The acquisition parameter adjustment instructions are fed back to the multispectral image acquisition and adaptive registration module to form a closed-loop feedback control mechanism.

2. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The multispectral image acquisition and adaptive registration module includes: A multispectral camera unit is used to simultaneously acquire visible light images, near-infrared images, and thermal infrared images of the area where the warm needle acupuncture is applied; A time synchronization unit, connected to the multispectral camera unit, is used to precisely synchronize and control the image acquisition time of each spectral channel. The feature point detection unit, connected to the time synchronization unit, is used to detect key feature points in each spectral image based on the scale-invariant feature transformation algorithm. The registration calculation unit, connected to the feature point detection unit, is used to calculate the spatial transformation matrix between different spectral images based on the key feature points, thereby completing image registration.

3. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The spectral fusion and deep feature extraction module includes: Principal component analysis unit is used to perform principal component analysis on the registered multispectral image set and extract principal component features; A multi-scale fusion unit, connected to the principal component analysis unit, is used to construct a multi-scale Gaussian pyramid and perform weighted fusion of principal component features at different scales. A deep feature encoding unit, connected to the multi-scale fusion unit, is used to extract spectral-spatial depth features of the fused spectral image based on a convolutional neural network.

4. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The organizational-level response analysis and depth estimation module includes: A depth estimation network unit is used in a deep learning network based on an encoder-decoder architecture to estimate the tissue depth distribution from the depth feature map; The tissue stratification unit, connected to the depth estimation network unit, is used to divide the subcutaneous tissue into the epidermis, dermis, subcutaneous fat layer and muscle fascia layer based on the tissue depth distribution map and physiological prior knowledge. The thermal response calculation unit, connected to the tissue layering unit, is used to calculate the temperature change rate, heat energy accumulation, and response time constant of each tissue layer.

5. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The closed-loop feedback control and visualization output module includes: The evaluation index calculation unit is used to calculate the evaluation indexes of tissue thermal energy penetration depth and response intensity based on the multi-level tissue response data. A three-dimensional visualization unit, connected to the evaluation index calculation unit, is used to generate a three-dimensional visualization depth response map; A deviation analysis unit, connected to the evaluation index calculation unit, is used to calculate the deviation between the evaluation index and the preset treatment target; The parameter optimization unit, connected to the deviation analysis unit, is used to generate acquisition parameter adjustment instructions and moxibustion parameter optimization suggestions based on the deviation.

6. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The visible light image wavelength range of the multispectral image set is 400nm to 700nm, the near-infrared image wavelength range is 700nm to 1400nm, and the thermal infrared image wavelength range is 8μm to 14μm.

7. The multispectral fused depth of warm needling and tissue response monitoring system of claim 1, wherein, The principal component analysis algorithm extracts 3 to 8 principal components, and the cumulative variance contribution rate retained is not less than 95%.

8. The multispectral fusion-based warm needling moxibustion depth and tissue response monitoring system according to claim 1, characterized in that, The depth estimation network adopts the U-Net architecture, with the encoder consisting of 5 levels of downsampling layers and the decoder consisting of 5 levels of upsampling layers, and skip connections are set between the encoder and the decoder.

9. The multispectral fusion-based warm needling moxibustion depth and tissue response monitoring system according to claim 1, characterized in that, The calculation of the tissue thermal penetration depth is based on a heat conduction model, and the response intensity evaluation index includes the peak temperature gradient, the rate of thermal energy accumulation, and the tissue response uniformity.

10. The multispectral fusion-based warm needling moxibustion depth and tissue response monitoring system according to claim 1, characterized in that, The feedback cycle of the closed-loop feedback control mechanism is 1s to 5s. The parameter adjustment instructions include exposure time adjustment, gain adjustment and sampling frequency adjustment. The moxibustion parameter optimization suggestions include moxibustion temperature suggestions, moxibustion time suggestions and acupuncture depth suggestions.

Images