Subcutaneous injection site detection and recommendation system based on intelligent algorithm

Abstract

The invention discloses a subcutaneous injection site detection and recommendation system based on an intelligent algorithm, and aims to improve the safety of long-term subcutaneous injection and the drug absorption efficiency and reduce the risk of related complications such as subcutaneous hemorrhage, subcutaneous induration and subcutaneous fat hyperplasia at the same time. Comprising a subcutaneous tissue detection module for acquiring subcutaneous vascular distribution, fat layer thickness and induration state of a target part through infrared imaging and ultrasonic technologies and generating subcutaneous tissue characteristic data; the data processing module analyzes the feature data by using an image recognition algorithm, and recognizes a blood vessel dense region, induration and a part where fat hyperplasia has occurred; the injection part recommendation module is used for evaluating an analysis result of the data processing module based on a machine learning algorithm in combination with personalized injection historical data of the patient, and generating an optimal injection part recommendation scheme avoiding a risk area; the intelligent imaging and data analysis technology is utilized, the subcutaneous injection part selection accuracy is improved, and the patient comfort and the treatment effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a subcutaneous injection site detection and recommendation system based on an intelligent algorithm. Background Art

[0002] Subcutaneous injection is an important method of drug delivery and is widely used in the treatment of chronic diseases such as diabetes and rheumatoid arthritis. Traditional subcutaneous injection methods rely mainly on the experience of patients or medical staff to select the site, usually based on anatomical guidelines to avoid areas with dense blood vessels and areas with existing complications. Some patients use palpation to identify areas of induration or fatty hyperplasia to reduce the occurrence of adverse reactions. In addition, some auxiliary devices have emerged in recent years, such as needle puncture depth control devices and local skin scanning devices, to help optimize the injection effect.

[0003] Existing technologies have certain problems with subcutaneous injection site selection. First, manual or empirical judgment can easily lead to errors, and patients may not be able to accurately identify the appropriate injection area, increasing the risk of subcutaneous bleeding, nodules, and fat hyperplasia. Second, most existing auxiliary equipment lacks intelligent analysis capabilities and cannot be dynamically adjusted based on individual long-term injection data, resulting in patients being unable to obtain personalized injection recommendations. In addition, the lack of systematic data recording and analysis methods makes long-term optimization and management of injection sites more difficult.

[0004] In order to solve the above problems, the present invention proposes a subcutaneous injection site detection and recommendation system based on intelligent algorithm. Summary of the Invention

[0005] This application provides a subcutaneous injection site detection and recommendation system based on intelligent algorithms to improve the safety of long-term subcutaneous injections and the efficiency of drug absorption.

[0006] This application provides a subcutaneous injection site detection and recommendation system based on an intelligent algorithm, including: a subcutaneous tissue detection module configured to detect the subcutaneous blood vessel distribution, fat layer thickness, and induration status of a target site through infrared imaging and ultrasonic technology, and generate corresponding subcutaneous tissue characteristic data; a data processing module, in communication with the subcutaneous tissue detection module, configured to receive the subcutaneous tissue characteristic data and identify vascular dense areas, indurations, and areas where fat hyperplasia has occurred through an image recognition algorithm; An injection site recommendation module analyzes the recognition results of the data processing module based on a machine learning algorithm and combines the patient's personalized injection history data to generate personalized injection site recommendations that avoid areas with dense blood vessels, nodules, and areas where fat hyperplasia has occurred, thereby reducing the risk of complications related to subcutaneous bleeding, subcutaneous nodules, and subcutaneous fat hyperplasia, while optimizing drug absorption efficiency. The user interaction module is configured to present the personalized injection site recommendation to the user and record the site and effect data of each injection to support long-term monitoring and treatment adjustment.

[0007] Furthermore, the subcutaneous tissue detection module includes an infrared imaging unit and an ultrasonic imaging unit, wherein the infrared imaging unit is configured to collect thermal images of the target part under two or more different infrared bands, and highlight the direction of blood vessels and thermal abnormality areas through image superposition and contrast enhancement processing; the ultrasonic imaging unit is configured to emit ultrasonic signals of different frequencies in sequence, and determine the thickness of the fat layer and the boundary of the abnormal high-density area based on the change in echo intensity at each frequency; the images collaboratively acquired by the infrared imaging unit and the ultrasonic imaging unit are processed through spatial coordinate alignment to form subcutaneous tissue feature data including blood vessel distribution, fat layer thickness and abnormal tissue boundary information.

[0008] Furthermore, the data processing module includes an image fusion receiving unit, a fusion image deconstruction unit and a feature data structure generation unit, wherein: The image fusion receiving unit is configured to receive a fusion image output by the subcutaneous tissue detection module after spatial coordinate alignment processing, wherein the fusion image includes the vascular thermal response distribution, fat layer thickness information and abnormal tissue boundary in the target area; The fused image deconstruction unit is configured to deconstruct image signals corresponding to different tissue types in the fused image by a layered analysis method, including extracting a thermal gradient trend layer associated with blood vessels, an echo amplitude distribution layer associated with fat layers, and a high-density boundary marker layer associated with induration in the fused image; The feature data structure generating unit is configured to generate first subcutaneous tissue feature data in a unified format based on the multi-layer image information extracted by the fused image deconstruction unit, including the spatial distribution, boundary range and tissue attribute labels of various risk areas in a unified coordinate system.

[0009] Furthermore, the injection site recommendation module constructs a scoring function based on the support vector machine model to quantitatively score the injection suitability of each candidate injection point. The scoring function is defined as the following formula 1: in, is the number of historical sample points used to train the support vector machine model, each of which has a labeled feature vector and spatial coordinates; represents the injection suitability score; Indicates candidate injection sites The corresponding eigenvector; For the The support vector weights corresponding to the training samples are obtained through support vector machine model training; For the training set The feature vector of each sample; is the kernel width parameter, which is used to adjust the mapping scale of the feature space; is the model bias term; is the position weight function; wherein the feature vector is expressed by the following formula 2: in, Candidate injection sites The vascular density coefficient at a point is calculated based on the number of vascular thermal responses detected per unit area in a local window centered at the point in the fused image; Candidate injection sites The vertical thickness of the fat layer was derived from the depth measurement of the fat tissue layer in ultrasound images; Candidate injection sites The proportion of abnormal tissue in the surrounding area is defined as The percentage of the area identified as induration, scar or fat hyperplasia in the image to the entire neighborhood area within the center; Candidate injection sites The historical injection frequency is provided by the user injection record system and is defined as the ratio of the number of times the point is used in the user's historical injection distribution to the total number of injections; Indicates candidate injection points The corresponding tissue reaction level label is obtained by evaluating the user's feedback data within 48 hours after the injection at the same point. The feedback data includes the area of redness and swelling, subjective pain score, and the amplitude of the change in local bioelectrical impedance; Position weight function Calculated by the following formula 3: in, Candidate injection sites With training sample points Euclidean distance in body surface space; is the structural risk adjustment factor; The candidate points with positive injection suitability scores are determined as injectable areas.

[0010] Furthermore, the infrared imaging unit is configured to acquire time series thermal images in a plurality of continuous infrared bands, and extract dynamic blood flow characteristics by analyzing the temperature gradient change rate, so as to distinguish the direction of veins and arteries.

[0011] Furthermore, the ultrasonic imaging unit is configured to construct a tissue elasticity map based on the echo phase difference, which is used to assist in identifying potential nodule areas that have not formed obvious density differences in the early stage, thereby improving the early detection capability of abnormal tissue and enhancing the accuracy of risk area identification.

[0012] Furthermore, the fused image deconstruction unit is configured to extract composite risk areas based on the pixel-level spatial overlapping relationship between each layer, and assign a high-risk weight label to the intersection area of vascular thermal response and abnormal density boundary; the feature data structure generation unit is configured to encode various risk areas in layers according to risk levels, and establish time-series difference data between consecutive frame images for analyzing the dynamic change trend of local tissue status, thereby improving the timeliness and accuracy of injection site risk assessment.

[0013] Furthermore, the user interaction module includes a visual feedback interface unit, an injection feedback collection unit and an individual behavior recognition unit, wherein the visual feedback interface unit is configured to spatially mark the personalized injection site recommendation plan in the form of a body surface topology map in the display interface, and use color coding to distinguish injectable areas, contraindicated areas and priority recommended areas; the injection feedback collection unit is configured to guide the user to enter the injection location, drug type and tissue reaction within 48 hours after the injection is completed, including the area of redness and swelling, pain level and tissue tactile changes, and automatically associate and store the feedback information with the recommended injection point; the individual behavior recognition unit identifies the patient's compliance, tendency and injection habits in actual use based on the user's long-term injection data and feedback pattern, and adjusts the weight parameters of future recommendation strategies to achieve cross-user personalized modeling and adaptive optimization.

[0014] Furthermore, the user interaction module is configured to generate a user-specific injection management cycle, and dynamically lock the recommended injection site plan or provide risk upgrade prompts based on the injection results and complication reports input by the user. The user interaction module includes a cycle guidance unit, which is used to generate an injection rotation sequence and a minimum injection interval between different anatomical areas on the body surface based on the recommended areas identified by the system, and prompt the user to perform the injection as planned in a graphical calendar format; the complication perception unit is used to receive local erythema, persistent nodules, and infection signs actively input by the user or identified by the system, and upon detecting that a specific risk indicator exceeds a preset threshold, trigger the temporary closure of the corresponding injection area, risk warning, and doctor intervention prompt functions, thereby realizing a real-time interactive response mechanism based on the dynamic changes of complications.

[0015] Furthermore, the user interaction module includes a body surface guidance calibration unit, which is configured to, after generating a personalized injection site recommendation plan, calibrate the spatial matching relationship between the user's current position and the recommended injection point in combination with the system positioning component or image re-recognition function, and prompt the user to complete precise injection positioning through a multi-modal method of screen guidance, vibration feedback, and image overlay; the body surface guidance calibration unit is also configured to, after the user manually confirms or touches the selected injection point, capture the injection site image through the camera, perform feature alignment with the original recommendation layer, and determine whether the actual injection point deviates from the recommended area. If the deviation exceeds the system set tolerance, a re-positioning prompt and a deviation risk prompt are triggered.

[0016] This application has the following beneficial technical effects: (1) By using infrared imaging and ultrasonic technology to detect subcutaneous blood vessel distribution, fat layer thickness and induration status in real time, and combining image recognition algorithms to accurately identify areas with dense blood vessels, induration and fat hyperplasia, the risk of misjudgment is reduced, thereby effectively reducing the incidence of complications such as subcutaneous bleeding, subcutaneous induration and fat hyperplasia, and improving injection safety. (2) By using machine learning algorithms combined with patients' personalized injection history data, the optimal injection area is intelligently analyzed to avoid induration and fat hyperplasia areas, thereby improving the uniform distribution and absorption efficiency of drugs in subcutaneous tissue, thereby reducing blood sugar fluctuations or other unstable drug effects and improving treatment effects. (3) By long-term recording of each injection site and effect data, combined with intelligent analysis models to dynamically optimize individual injection strategies, patients can obtain more accurate personalized injection guidance, avoid local tissue damage caused by long-term use of the same area, and improve the feasibility of long-term subcutaneous injection treatment and patient compliance. (4) A user interaction module is used to present the recommended injection site through a visual interface, so that patients can intuitively understand the most suitable injection location. At the same time, it has data storage and long-term monitoring functions, which reduces the trouble of patients relying on experience and judgment, improves convenience and operational efficiency, and makes home and clinical use more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of a subcutaneous injection site detection and recommendation system based on an intelligent algorithm provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0018] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0019] The first embodiment of this application provides a subcutaneous injection site detection and recommendation system based on intelligent algorithms. Figure 1 , which is a schematic diagram of the first embodiment of this application. Figure 1 The first embodiment of the present application provides a subcutaneous injection site detection and recommendation system based on an intelligent algorithm, which is described in detail.

[0020] The intelligent algorithm-based subcutaneous injection site detection and recommendation system includes a subcutaneous tissue detection module 101 , a data processing module 102 , an injection site recommendation module 103 and a user interaction module 104 .

[0021] The subcutaneous tissue detection module 101 is configured to detect the subcutaneous blood vessel distribution, fat layer thickness and induration state of the target part through infrared imaging and ultrasonic technology, and generate corresponding subcutaneous tissue characteristic data.

[0022] Subcutaneous tissue detection module 101 is used to perform real-time detection of the target subcutaneous injection site, acquiring characteristic data on subcutaneous vascular distribution, fat layer thickness, and induration status. This module combines infrared imaging and ultrasound technology to ensure detection accuracy and applicability, meeting the needs of patients undergoing long-term subcutaneous injections.

[0023] Infrared imaging technology exploits the temperature distribution characteristics of subcutaneous tissue, collecting skin surface temperature information using short-wave or medium-wave infrared sensors. Specific image processing algorithms are then used to enhance the contrast of vascular structures. To improve detection stability, the module employs active infrared illumination technology, using a short-wave near-infrared LED or laser source to illuminate the skin surface. This makes changes in infrared absorption rates in the vascular region more pronounced, resulting in clearer vascular images. Furthermore, the system incorporates an analytical model based on thermal diffusion characteristics to further enhance vascular structure detection by analyzing local temperature gradients and blood flow patterns. To reduce ambient light interference, the module can integrate a narrowband optical filter and implement background temperature compensation algorithms for correction. Furthermore, the infrared sensor's resolution and sampling frequency can be adjusted according to detection requirements. A high-resolution sensor with a resolution of 640×480 pixels or higher and a frame rate of 30Hz or higher is generally recommended for dynamic data acquisition to ensure real-time performance and accuracy.

[0024] Ultrasonic imaging technology uses a high-frequency ultrasound probe to image subcutaneous tissue. Using a 5MHz to 15MHz linear or phased array ultrasound probe, it can accurately measure subcutaneous fat thickness and identify the formation of induration. The probe consists of a piezoelectric transducer that transmits ultrasonic pulses and receives echo signals from the subcutaneous tissue. The echo signals undergo amplification, filtering, envelope detection, and digital signal processing to ultimately produce a high-resolution ultrasound image. To improve imaging quality, the module utilizes adaptive gain control (TGC) technology to maintain a good signal-to-noise ratio at all depths. It also combines Fourier transform and wavelet transform methods for echo signal analysis to extract subcutaneous tissue texture features, enhancing the detection of fat hyperplasia and induration. Furthermore, color Doppler mode can be used to monitor local blood flow and further identify areas of dense vascularization to avoid puncturing blood vessels and causing bleeding during injection.

[0025] This module integrates an image acquisition unit and a data processing unit to achieve efficient data acquisition and preprocessing. The image acquisition unit is responsible for simultaneously acquiring infrared and ultrasonic data, and performing temporal synchronization and spatial alignment to ensure that the two imaging technologies can provide complementary information. The data processing unit uses an efficient data fusion algorithm to overlay infrared and ultrasonic images through image registration technology, and adopts multi-scale edge detection and region segmentation algorithms to extract vascular structure, fat layer thickness, and nodule areas respectively. To further improve detection accuracy, the unit can use deep learning-based feature extraction networks, such as U-Net or ResNet, to automatically segment and label images, ensuring the accuracy and stability of data analysis.

[0026] The hardware components of this module can be designed to be portable or wearable to suit different usage scenarios. For example, a portable version could utilize a handheld scanning device with an integrated high-definition LCD screen to display test results in real time, and transmit the data to the data processing module 102 via wireless transmission methods (such as Wi-Fi or Bluetooth). A wearable version could be integrated into a smart patch or wristband, enabling long-term patient monitoring. Historical data could be stored in the cloud to support long-term trend analysis and personalized treatment adjustments.

[0027] The design of this module ensures accurate acquisition and real-time analysis of subcutaneous tissue data, enabling the system to provide high-precision injection site assessment, thereby reducing injection-related complications, improving drug absorption efficiency, and optimizing the effects of long-term subcutaneous injection therapy.

[0028] Furthermore, the subcutaneous tissue detection module includes an infrared imaging unit and an ultrasonic imaging unit, wherein the infrared imaging unit is configured to collect thermal images of the target part under two or more different infrared bands, and highlight the direction of blood vessels and thermal abnormality areas through image superposition and contrast enhancement processing; the ultrasonic imaging unit is configured to emit ultrasonic signals of different frequencies in sequence, and determine the thickness of the fat layer and the boundary of the abnormal high-density area based on the change in echo intensity at each frequency; the images collaboratively acquired by the infrared imaging unit and the ultrasonic imaging unit are processed through spatial coordinate alignment to form subcutaneous tissue feature data including blood vessel distribution, fat layer thickness and abnormal tissue boundary information.

[0029] In this embodiment, the subcutaneous tissue detection module is further refined to consist of an infrared imaging unit and an ultrasonic imaging unit, which respectively obtain tissue information of different dimensions of the target area, and generate clinically significant subcutaneous tissue feature data through collaborative processing for subsequent data processing and injection site recommendation by the system.

[0030] The infrared imaging unit is used to obtain thermal distribution images of the skin surface. It is configured to be able to image in two or more different infrared bands. For example, short-wave infrared (and medium-wave infrared) bands can be selected. By adopting a multi-band infrared sensor or switching different band filters on a single sensor, multiple sets of thermal images of the target skin area are collected. Because the thermal conductivity characteristics of subcutaneous blood vessels are different from those of surrounding tissues, the thermal contrast exhibited when imaging in different bands has certain differences. The system performs superposition processing on these images. The superposition processing can adopt image pixel-level averaging, weighted fusion, or maximum value to enhance the overall contrast of the image. Subsequently, image contrast enhancement processing is performed, such as local contrast stretching, histogram equalization, or edge enhancement filtering, to make the thermal response contour of the subcutaneous blood vessels clearer, facilitating the extraction of their spatial distribution characteristics. In addition, this infrared imaging process does not require direct contact with the skin and is suitable for non-invasive use in home or clinical environments.

[0031] The ultrasound imaging unit is configured to transmit ultrasonic signals of varying frequencies through one or more transducer arrays. For example, pulse waves at frequencies such as 5 MHz, 10 MHz, and 15 MHz can be sequentially emitted, and the echo signal intensity corresponding to each frequency can be recorded. In ultrasound imaging, the echo intensity of tissue varies with differences in its acoustic impedance and scattering properties. Low-frequency signals can penetrate deeper tissue layers at the expense of resolution, while high-frequency signals are suitable for high-resolution imaging of shallow layers. Therefore, by comparing echo images at different frequencies, the system analyzes the depth and uniformity of the fat layer and identifies the location and boundaries of abnormally high-density tissue (such as nodules or fibrotic nodules). Boundaries can be determined based on abrupt changes in the echo signal at specific depths. Specifically, when the echo amplitude significantly increases or decreases over a short distance, the system identifies a possible tissue interface.

[0032] Because infrared and ultrasound images have different acquisition mechanisms, their spatial resolution and viewing angles differ. Therefore, after processing both images, this system achieves image fusion through spatial coordinate alignment. Specifically, image geometric correction is first achieved using standard detection postures or markers set by the system, so that the two types of images have the same reference coordinate system. Subsequently, image registration techniques such as affine transformation or bilinear interpolation are used to overlay the infrared image on the ultrasound image. This ultimately creates fused image data, which simultaneously annotates the subcutaneous blood vessel direction, fat layer thickness, and abnormal tissue boundary information. This data is then output as structured data and transmitted to the data processing module as subcutaneous tissue feature data.

[0033] The above technical solution achieves high-precision identification of multiple types of subcutaneous tissue structures by combining two complementary imaging technologies, infrared and ultrasound, and ensures the spatial consistency of data through image registration.

[0034] Furthermore, the infrared imaging unit is configured to acquire time series thermal images in a plurality of continuous infrared bands, and extract dynamic blood flow characteristics by analyzing the temperature gradient change rate, so as to distinguish the direction of veins and arteries.

[0035] In this embodiment, the infrared imaging unit not only possesses multi-band imaging capabilities but is also configured to perform time-series image acquisition and dynamic analysis processing to further improve the accuracy of vascular identification, particularly important for functionally distinguishing between veins and arteries. This infrared imaging unit utilizes a detector that supports multiple mid- and far-infrared bands (e.g., 3–5µm and 8–12µm). Combined with a filter switching mechanism or multispectral thermal sensor, this unit can simultaneously or rapidly sequentially acquire thermal images of the target body surface area within the same field of view, thereby obtaining thermal distribution images in multiple bands.

[0036] In practice, the infrared imaging unit continuously samples the target injection area for a short period of time at a certain frame rate (e.g., 5–20 frames per second), forming a time series of thermal images. In each frame, the area corresponding to the blood vessel typically exhibits a higher or lower infrared radiation temperature than the surrounding tissue. Especially in areas of active blood flow, the temperature response undergoes subtle but detectable dynamic fluctuations with blood pulsation. The system calculates the time-dependent rate of change of the temperature gradient at each pixel—that is, the temperature difference at a pixel in adjacent time frames divided by the time interval—to determine the intensity of the temperature change at that location.

[0037] By extracting the continuous spatial distribution of these temperature gradient change rates, a dynamic thermal response map is constructed, and blood flow activity characteristics are analyzed accordingly. Because blood flow in arteries is faster and pressure fluctuations are more significant, the amplitude and frequency of temperature fluctuations in thermal images are generally higher than those in veins. Therefore, the system can distinguish the dynamic performance of arteries and veins in thermal image sequences by setting thresholds, spectral analysis, or wavelet transforms. After image enhancement and vascular direction extraction, the system can clearly mark the arterial and venous paths, providing more detailed and accurate anatomical reference information for safe avoidance of subsequent injection sites.

[0038] The key to this implementation lies in leveraging the temporal microdynamics of thermal images, extracting information about vascular activity through the rate of change of temperature gradients, rather than relying solely on the static thermal signature of a single frame. This non-contact, non-invasive, and highly sensitive approach can be combined with subsequent ultrasound imaging and tissue recognition data for analysis, improving the system's ability to identify high-risk injection areas and ensuring the accuracy and safety of the injection process.

[0039] Furthermore, the infrared imaging unit is configured to perform frequency domain analysis on the time series thermal images by Fourier transform or wavelet decomposition, and identify the arterial area in combination with the pulse physiological frequency band.

[0040] In this embodiment, the infrared imaging unit is constructed to have multi-band thermal imaging and continuous time series acquisition capabilities, and introduces a time-frequency joint analysis method to extract temperature change patterns directly related to the dynamic characteristics of blood flow, thereby realizing automatic identification and separation of arteries and veins. Specifically, the infrared imaging unit includes a thermal detector array that supports synchronous or interleaved acquisition in two or more mid- and far-infrared bands, such as capturing images of the same target area in the 3–5μm and 8–12μm bands. Compared with single-band thermal imaging technology, multi-band information can enhance temperature contrast and improve the separability between vascular thermal response signals and background tissue, which is particularly suitable for distinguishing deep and shallow blood vessels under complex surface structures.

[0041] To capture thermal trends caused by blood flow, the infrared imaging unit continuously captures the same target injection area for short periods of time (e.g., 2 to 5 seconds) at a frequency of 10 to 30 frames per second, generating a multi-band, time-spanning thermal image sequence. After receiving this thermal image sequence, the system performs a time-domain temperature gradient rate analysis on each spatial pixel. This involves performing a first-order difference on the pixel's temperature values in consecutive frames, then smoothing the temperature using a sliding window to obtain a temperature trajectory over the entire observation period. This trajectory typically manifests as a fluctuating curve with a significant pulsation period in the arterial region, while a low-frequency, slowly varying, or near-steady-state signal appears in the venous region.

[0042] The system then performs a Fourier transform or wavelet decomposition on this temperature trajectory to obtain the power spectral density of each pixel in a specific frequency band. A bandpass filter is then set based on the physiological heart rate range (e.g., 0.8–1.5 Hz) to isolate high-frequency signal components synchronized with the pulse from the spectrum, indicating arterial location. Low-frequency, stable signals are more likely to correspond to veins or non-vascular tissue. Furthermore, to improve spatial consistency, the system performs a two-dimensional spatial convolution operation on the frequency feature map, integrating information from neighboring pixels to form a thermal dynamic response map. A complete vascular path map is then generated using threshold determination and a connected region extraction algorithm.

[0043] To further enhance the system's adaptability to diverse body shapes, skin tones, and basal body temperature variations, the present invention also designs a blood flow behavior recognition mechanism based on standardized dynamic template matching. This mechanism, based on a temperature variation template representing typical arterial and venous regions, utilizes a dynamic time warping (DTW) algorithm to perform temporal alignment and matching scoring on the actual collected temperature waveforms. This allows for the identification of atypical or weak vascular signals, enhancing robustness in the recognition of marginal and flattened vascular structures.

[0044] After all analysis steps are completed, the system overlays the arterial trajectory from the heat map sequence onto the fused image as a highlighted path marker. It also provides risk warnings for venous areas, allowing the subsequent injection site recommendation module to avoid high-risk areas when making decisions. Compared to traditional infrared vascular imaging, which extracts static thermal features based solely on brightness contrast, this implementation achieves blood flow behavioral pattern recognition by constructing a joint time-frequency analysis model. This approach is particularly suitable for deep arteries or special physiological structures that are difficult to identify using static heat maps, providing technical support for the precise risk management of long-term, multi-point injections.

[0045] In summary, the implementation of this infrared imaging unit not only relies on existing multi-band thermal imaging hardware and basic image processing procedures, but also combines multi-dimensional information processing methods such as time series analysis, frequency domain feature extraction and behavior matching recognition. The technical solution is significantly superior to the existing thermal imaging-assisted injection system in terms of vascular classification accuracy, adaptability and system linkage.

[0046] Furthermore, the ultrasonic imaging unit is configured to construct a tissue elasticity map based on the echo phase difference, which is used to assist in identifying potential nodule areas that have not formed obvious density differences in the early stage, thereby improving the early detection capability of abnormal tissue and enhancing the accuracy of risk area identification.

[0047] In this embodiment, the ultrasound imaging unit not only possesses traditional B-mode grayscale imaging capabilities but also integrates tissue elasticity imaging based on echo phase difference, enabling sensitive detection of early abnormal tissue areas, particularly potential indurations that have not yet demonstrated significant ultrasound echo density differences. This elasticity imaging process monitors the subtle deformations of the target tissue caused by mechanical perturbations. By comparing the ultrasound echo phase differences at each pixel before and after the deformation, the local rigidity of the tissue is reflected, indirectly inferring the elastic distribution of the tissue.

[0048] In practice, the ultrasound imaging unit uses structurally controllable, low-intensity mechanical perturbations, such as slight compression applied by the probe itself or system-controlled periodic compression excitation, to induce a certain degree of tissue elastic response in the target area. The system continuously acquires ultrasound echo data from the area at a high frame rate (e.g., dozens of frames per second) and performs differential calculations on the phase changes of the echo signals between adjacent frames to obtain the phase displacement of each pixel. Subsequently, the strain distribution map of the entire imaging area, namely the tissue elasticity map, is reconstructed by mapping the phase displacement map to the spatial position.

[0049] In this elastogram, relatively rigid areas experience smaller phase changes due to compression, while softer tissues experience larger phase changes. Induration, a pathological change caused by localized tissue fibrosis or chronic inflammation, typically has significantly higher elasticity than the surrounding normal subcutaneous tissue, resulting in a low-strain, strongly localized signal characteristic on the elastogram. Traditional B-mode ultrasound images may have difficulty identifying such abnormalities, especially in the early stages before significant acoustic density contrast has formed. However, elastograms based on phase difference reconstruction can reveal subtle abnormalities in tissue structural rigidity in advance, providing early warning of potential induration areas.

[0050] In the subsequent image processing module, the system spatially aligns the elastogram with the conventional echo intensity image and fuses it with the high-density tissue distribution layer to generate a composite image with a distribution of induration risk levels. This image data is further incorporated into the primary subcutaneous tissue feature data structure, which is then used by the subsequent injection site recommendation module to assess the potential tissue risk of each candidate injection area.

[0051] By introducing echo phase difference analysis into elastic imaging, the described embodiment significantly improves the ability to identify local pathological tissue changes. It is particularly suitable for the identification of chronic micronodules, which is common in long-term injection patients. It solves the problem that traditional identification methods that rely on tissue density are difficult to identify hidden lesions at an early stage.

[0052] The data processing module 102 is in communication with the subcutaneous tissue detection module and is configured to receive the subcutaneous tissue characteristic data and identify blood vessel dense areas, nodules and areas where fat hyperplasia has occurred through an image recognition algorithm.

[0053] The data processing module 102 receives the subcutaneous tissue feature data collected by the subcutaneous tissue detection module 101 and analyzes it to identify areas with dense blood vessels, induration, and areas with fatty hyperplasia. This module includes a data receiving unit, a preprocessing unit, an image recognition unit, and a data fusion unit to ensure data integrity and accuracy and improve the accuracy of target area recognition.

[0054] The data receiving unit is responsible for acquiring infrared imaging data, ultrasonic imaging data, and related physiological characteristic information from the subcutaneous tissue detection module 101. Because the raw data formats of infrared and ultrasonic imaging differ, the data receiving unit utilizes multi-channel data synchronization technology to ensure that data from different imaging sources is processed using the same time base. Furthermore, the unit supports wireless transmission protocols (such as Wi-Fi and Bluetooth) or wired transmission interfaces (such as USB and serial ports) to meet the connection requirements of different devices.

[0055] The preprocessing unit performs preliminary processing on the received image and signal data, including noise removal, contrast enhancement, edge detection, and feature extraction. Infrared imaging data undergoes temperature calibration to eliminate the effects of ambient temperature changes on imaging. High-pass filtering and threshold segmentation methods are used to enhance the visibility of vascular structures. Ultrasound data undergoes time-domain filtering, frequency-domain enhancement, and artifact removal to ensure a clear presentation of fat layer thickness and induration areas. Furthermore, to improve data consistency, the preprocessing unit employs image registration technology to spatially align data from different sources for subsequent comprehensive analysis.

[0056] The image recognition unit automatically analyzes preprocessed data using a combination of deep learning algorithms and traditional computer vision methods. Identification of densely vascularized areas is based on a convolutional neural network (CNN). This model learns the characteristics of different subcutaneous vascular patterns from a training dataset and automatically segments these areas using a morphological processing algorithm. Identification of induration relies on ultrasound echo texture analysis, which determines the location of potential indurations by statistically analyzing the energy distribution and tissue reflectivity characteristics of the echo signal. Furthermore, detection of areas of fatty hyperplasia utilizes a deep learning model for image semantic segmentation, combined with edge detection and region growing algorithms, to determine the extent and severity of the induration.

[0057] The data fusion unit comprehensively processes infrared imaging, ultrasonic imaging, and analysis results to improve recognition reliability and accuracy. This unit utilizes a multimodal data fusion approach, combining feature-level fusion and decision-level fusion to ensure comprehensive analysis of each detection indicator. Feature-level fusion methods include principal component analysis (PCA) and deep feature extraction, which transform the features of infrared and ultrasonic imaging data to extract complementary information. Decision-level fusion methods utilize a weighted voting mechanism to combine multiple recognition results to provide the optimal recognition judgment. Furthermore, the data fusion unit can perform adaptive adjustments based on historical data, allowing the system to continuously optimize recognition accuracy over the long term.

[0058] The overall design of the data processing module 102 ensures efficient analysis and intelligent identification of subcutaneous tissue detection data, so that areas with dense blood vessels, nodules and fat hyperplasia can be accurately marked, providing reliable data support for subsequent personalized injection site recommendations.

[0059] Furthermore, the data processing module includes an image fusion receiving unit, a fusion image deconstruction unit and a feature data structure generation unit, wherein: The image fusion receiving unit is configured to receive a fusion image output by the subcutaneous tissue detection module after spatial coordinate alignment processing, wherein the fusion image includes the vascular thermal response distribution, fat layer thickness information and abnormal tissue boundary in the target area; The fused image deconstruction unit is configured to deconstruct image signals corresponding to different tissue types in the fused image by a layered analysis method, including extracting a thermal gradient trend layer associated with blood vessels, an echo amplitude distribution layer associated with fat layers, and a high-density boundary marker layer associated with induration in the fused image; The feature data structure generating unit is configured to generate first subcutaneous tissue feature data in a unified format based on the multi-layer image information extracted by the fused image deconstruction unit, including the spatial distribution, boundary range and tissue attribute labels of various risk areas in a unified coordinate system.

[0060] In this embodiment, the data processing module includes an image fusion receiving unit, a fusion image deconstruction unit, and a feature data structure generation unit, which work together in sequence to complete the reception, layered interpretation, and standardized output of the fusion image.

[0061] The image fusion receiving unit is used to receive the fused image output by the subcutaneous tissue detection module. The fused image is a composite image generated after the front-end infrared imaging unit and the ultrasound imaging unit respectively collect image data and complete the spatial coordinate alignment processing. The fused image is an image with multi-dimensional tissue information formed based on multiple imaging channels of the same target area. It contains at least three core tissue features: the first is the vascular thermal response distribution, which comes from the high thermal gradient structure formed at the blood vessel position due to the tiny temperature difference changes on the human skin surface in infrared imaging; the second is the fat layer thickness information, which is calculated based on the attenuation pattern of the ultrasound echo at different depths and reflects the longitudinal thickness changes of the subcutaneous fat tissue; the third is the abnormal tissue boundary, which is mainly manifested as a high-reflection area with drastic density changes and clear edges in the ultrasound image, corresponding to the actual hard nodules or fibrotic tissue.

[0062] The function of the fusion image deconstruction unit is to separate the layers and analyze the target structure of the received composite image. The unit uses a hierarchical analysis method to identify and classify the image content. For the blood vessel part, the unit extracts the direction of the continuous distribution area of the thermal gradient in the thermal map, and forms a thermal response layer representing the direction of the blood vessel; for the fat layer, the unit extracts the morphological boundary of the continuous low echo area from the ultrasound channel, and forms a grayscale distribution layer representing the fat thickness; for the nodule or abnormal density tissue, the unit locates the area with concentrated high echo intensity and obvious edges, and forms a high-density tissue boundary layer. Each layer retains the original Figure 1 The consistent spatial coordinate information can be used to achieve unified spatial recognition and position marking in the subsequent processing.

[0063] Based on the multi-layer image information obtained through the deconstruction, the feature data structure generation unit generates first subcutaneous tissue feature data in a unified format. This data structure not only contains the image information of each tissue structure, but also further extracts its geometric properties and spatial distribution parameters in a unified coordinate system. Each risk area (such as vascular-dense areas, fatty hyperplasia areas, and induration areas) is annotated with its boundary contour, center location, area, surrounding tissue type, and risk level attributes. This feature data is transmitted in a structured form to the injection site recommendation module, providing basic data support for subsequent intelligent analysis and recommendations.

[0064] Furthermore, the fused image deconstruction unit is configured to extract composite risk areas based on the pixel-level spatial overlapping relationship between each layer, and assign a high-risk weight label to the intersection area of vascular thermal response and abnormal density boundary; the feature data structure generation unit is configured to encode various risk areas in layers according to risk levels, and establish time-series difference data between consecutive frame images for analyzing the dynamic change trend of local tissue status, thereby improving the timeliness and accuracy of injection site risk assessment.

[0065] In this embodiment, after completing the layered analysis of the fused image, the fused image deconstruction unit further constructs composite risk regions based on the pixel-level spatial overlap between the layers, thereby improving the comprehensive identification of high-risk injection areas. A fused image typically includes information on multiple layers, such as vascular thermal response layers, fat layer thickness layers, and abnormally high-density tissue boundary layers. Each layer has a two-dimensional spatial structure aligned with a unified coordinate system. The system performs pixel-by-pixel alignment and comparison of these layers to identify overlapping areas on the image matrix that simultaneously meet two or more risk characteristics.

[0066] Specifically, when an area appears as a high-temperature gradient region in the thermal response layer (usually referring to dense blood vessels or obvious dynamic blood flow) and simultaneously exhibits a high-density edge signal in the density boundary layer (such as the boundary outline of a nodule, scar, or fatty hyperplasia), the system determines that the area is a composite risk area. Such areas should be avoided during actual injections because they carry the potential concurrent risks of vascular damage and local drug absorption disorders. To this end, the fused image deconstruction unit assigns these areas higher weight labels to guide subsequent injection site scoring and avoidance logic.

[0067] After extracting the composite risk, the feature data structure generation unit uses the identification results to hierarchically encode each risk area according to its risk level. This encoding can employ an integer rating system (e.g., 0 represents low risk, 1 represents medium risk, and 2 represents high risk). The data structure also retains the spatial extent, boundary outline, and source layer information of the corresponding area in a unified coordinate system. This risk level encoding is not only used for static decision-making but can also be combined with image sequences for temporal analysis.

[0068] To enhance its ability to perceive the evolution of injection site conditions, the system further establishes a timeline differential channel between consecutive frame fusion images. By comparing changes in the characteristic values of each layer at the same location across different time frames, such as increased fat thickness, enlarged nodule boundaries, or increased blood flow layer activity, the system can analyze whether there is a sustained risk increase trend in the local tissue over a short period of time. This dynamic trend information is stored in a feature data structure and used as a historical input feature in the scoring decision-making of subsequent machine learning models, further improving the system's ability to identify and accurately predict the evolution of abnormal tissues.

[0069] This approach leverages the spatial and temporal correlations of multi-layer image data, building upon spatial overlap analysis and temporal difference analysis mechanisms. This approach not only identifies risk areas based on their current static characteristics but also reflects their dynamic evolution. This approach is particularly well-suited for managing the progressively worsening tissue conditions of chronic injection sites, such as induration and scarring, and can significantly enhance the foresight and safety of personalized injection recommendations.

[0070] The injection site recommendation module 103 analyzes the recognition results of the data processing module based on a machine learning algorithm, and combines the patient's personalized injection history data to generate a personalized injection site recommendation plan that avoids areas with dense blood vessels, nodules, and areas where fat hyperplasia has occurred, so as to reduce the risk of complications related to subcutaneous bleeding, subcutaneous nodules, and subcutaneous fat hyperplasia, while optimizing drug absorption efficiency.

[0071] Injection site recommendation module 103 is primarily used to provide optimal subcutaneous injection site recommendations based on the analysis results of data processing module 102 and the patient's personalized injection history data. This module leverages machine learning algorithms to conduct in-depth analysis of subcutaneous tissue feature data, ensuring that recommended injection sites effectively avoid areas with dense blood vessels, nodules, and areas with established fat hyperplasia, thereby reducing the risks of subcutaneous bleeding, nodules, and subcutaneous fat hyperplasia while optimizing drug absorption efficiency.

[0072] This module first receives analytical data from data processing module 102, including information on subcutaneous vascularity, fat layer thickness, induration areas, and locations of adipose hyperplasia. Because subcutaneous tissue characteristics vary among individual patients, this module utilizes feature normalization methods to standardize subcutaneous data from different patients to ensure data consistency and minimize the impact of individual physiological differences on recommendation results. After data preprocessing, the module uses a specific region segmentation algorithm to divide the subcutaneous tissue into multiple potential injection areas and classify them according to their tissue characteristics.

[0073] The core of this module is decision analysis based on machine learning algorithms. First, models such as support vector machines (SVMs), random forests, or deep neural networks (DNNs) are used to train historical injection data to learn the long-term trends of different injection sites and their impact on drug absorption. During the training phase, the system uses supervised learning to leverage long-term patient data, including past injection sites, post-injection blood sugar control (for insulin users), and feedback from other physiological indicators, to categorize the advantages and disadvantages of different injection sites. The trained model is used to infer the optimal injection site based on new data, ensuring a highly personalized recommendation.

[0074] During the specific recommendation process, the module comprehensively considers multiple factors, including the state of subcutaneous tissue, historical injection distribution and its changing trends, individual patient characteristics (such as weight and subcutaneous fat thickness), drug absorption rate, and patient preferences. The system uses a weighted decision-making mechanism to score the suitability of different candidate injection sites. The scoring criteria include vascular density, uniformity of fat thickness, probability of nodule occurrence, historical frequency of use and its corresponding drug absorption feedback. During the final calculation process, the module uses Bayesian optimization or genetic algorithms to dynamically adjust the recommended parameters to ensure that the selected injection site meets physiological safety requirements and maximizes the stability of drug absorption.

[0075] To enhance the interpretability of the recommendation results, the module visualizes the finalized injection site and provides the basis for the recommendation, such as the site's vascularity, historical frequency of use, and other relevant characteristics. Furthermore, the system allows users to manually adjust recommendation parameters, for example, by choosing to avoid specific sites or prioritize certain areas for increased flexibility.

[0076] This module also features long-term learning and adaptive optimization. As patients continue to use the system, the system records information such as the injection site, drug dosage, and absorption effect (such as blood sugar fluctuations) after each injection. As this data accumulates, the module dynamically adjusts its recommendation strategy, enabling the system to adapt to long-term changes in individual patients, such as structural adjustments in subcutaneous tissue and changes in metabolic status. This dynamic optimization mechanism ensures that recommendations consistently meet the patient's optimal injection needs, improving the safety and effectiveness of long-term use.

[0077] The implementation of this module relies not only on algorithm optimization but also on hardware support. The system can be integrated into portable terminal devices, such as smart handheld devices and wearable medical devices, and linked to the user interaction module 104 using wireless communication technologies (such as Bluetooth or Wi-Fi), enabling convenient transmission of recommendation results to the user interface for viewing. For versions used by medical institutions, this module can be integrated with electronic medical record systems to achieve more comprehensive data management and medical decision support.

[0078] The design of this module ensures high accuracy and personalization of subcutaneous injection site recommendations, enabling patients to have a long-term safe and efficient injection experience, reduce the risk of complications, improve treatment compliance, and optimize drug absorption efficiency.

[0079] Furthermore, the injection site recommendation module constructs a scoring function based on the support vector machine model to The injection suitability is quantitatively scored, and the scoring function is defined as the following formula 1: in, represents the injection suitability score, where a value of +1 indicates that the point is suitable for injection, a value of −1 indicates that it is not suitable for injection, and 0 indicates that it is in a borderline uncertain state; Sign represents the sign function and is the final decision-making step in the entire scoring process. Its function is to convert the continuous numerical score, obtained by the system through the summation of weighted kernel functions, into a clear classification judgment. Specifically, the system first calculates the similarity between the feature vector and the support vector sample of the candidate injection point, and combines the position weight coefficient and the bias term to obtain a real value. The sign function determines the sign of this value: if the final result is greater than zero, the sign function outputs +1, indicating that the injection point is assessed as suitable for use; if the result is less than zero, the output is −1, indicating that injection is not recommended at this point; if the result is exactly zero, the output is 0, indicating that the point is at the scoring boundary, and the system can selectively include it in the candidate or conduct further analysis and processing.

[0080] Indicates candidate injection sites The corresponding eigenvector; is the number of historical sample points used to train the support vector machine model. Each sample point has a labeled feature vector and spatial coordinates, usually ranging from 50 to 200, depending on the richness of the historical injection data. Each training sample point contains a known label (indicating whether the injection effect at the point was good or bad in history) and the corresponding feature vector. ,These training samples come from the injection history records of real patients.

[0081] For the The support vector weights corresponding to the training samples are obtained through support vector machine model training; For the training set The feature vector of each sample; The width parameter of the kernel function is a key hyperparameter for adjusting the scale of the feature space mapping in the Gaussian radial basis function. The recommended initial value range is 0.3 to 1.2. The specific value is obtained through cross-validation to balance the risks of overfitting and underfitting. The model bias term is used to adjust the position of the hyperplane in the feature space and can be automatically learned during the SVM training process; is the position weight function; Eigenvector It is expressed using the following formula 2: in, Candidate injection sites The vascular density coefficient at a point is calculated based on the number of vascular thermal responses detected per unit area in a local window centered on the point in the fused image. Specifically, the number of vascular thermal responses detected in a circular neighborhood (e.g., 10 mm in diameter) centered on the point in the infrared image is counted and then divided by the neighborhood area to obtain the vascular density per unit area.

[0082] Candidate injection sites The vertical thickness of the fat layer was derived from the depth measurement of the fat tissue layer in ultrasound images; Candidate injection sites The proportion of abnormal tissue in the nearby area (e.g., a circular area with a diameter of 10 mm) is defined as The percentage of the area identified as induration, scar or fat hyperplasia in the image to the entire neighborhood area within the center; Candidate injection sites The historical injection frequency is provided by the user injection record system and is defined as the ratio of the number of times the point is used in the user's historical injection distribution to the total number of injections; Indicates candidate injection points The corresponding tissue reaction grade label is obtained by evaluating the user's feedback data within 48 hours after the injection at this point. The feedback data includes the area of redness and swelling, subjective pain score, and the amplitude of change in local bioelectrical impedance; specifically, it can be converted into three levels through a weighted scoring function (0 indicates a significant reaction, 1 indicates a mild reaction, and 2 indicates no adverse reaction).

[0083] Position weight function Calculated by the following formula 3: in, Candidate injection sites With training sample points Euclidean distance in surface space; this term is used to measure the similarity of spatial structure. The position weight function adjusts the spatial distance in a nonlinear manner to prevent distant training samples from having too strong an influence on the current point score.

[0084] It is a structural risk adjustment factor used to control the degree of spatial penalty. The recommended value is 0.05, which can be adjusted through model verification based on the actual data set.

[0085] The candidate points with positive injection suitability scores are determined as injectable areas.

[0086] After the scoring function is calculated, the system evaluates all candidate points and The point is determined as the injectable area.

[0087] The user interaction module 104 is configured to present the personalized injection site recommendation to the user and record the site and effect data of each injection to support long-term monitoring and treatment adjustment.

[0088] User Interaction Module 104 facilitates communication between the system and the user, ensuring patients can intuitively and conveniently access recommended subcutaneous injection sites. It also provides data recording and analysis capabilities to support long-term monitoring and personalized adjustments. This module is designed to enhance the user experience, enabling users to safely and efficiently perform subcutaneous injections based on system recommendations and leveraging historical data to optimize subsequent injection strategies.

[0089] This module first presents the analysis results obtained by the subcutaneous tissue detection module 101 and the data processing module 102 to the user through a graphical user interface, and visually displays the personalized injection site recommendations generated by the injection site recommendation module 103. Users can intuitively view the characteristic information of the subcutaneous tissue on the interface, including blood vessel distribution, fat thickness, nodule areas, and fat hyperplasia areas. The interface uses color to mark different types of tissue structures, such as using red to mark dense blood vessel areas, yellow to mark nodule areas, blue to mark fat hyperplasia areas, and highlighting the recommended safe injection areas in green. Users can adjust their injection site selection based on this information to avoid injecting into high-risk areas.

[0090] The system provides interactive options that allow users to manually adjust recommended injection sites. For example, patients can set certain areas as "prioritized" or "disabled," allowing the system to consider their preferences in future recommendations. For patients with specific needs, such as localized skin sensitivity or areas of injection pain, the module also allows users to mark these areas on the interface and provide feedback, allowing the system to dynamically adjust its strategy in subsequent recommendations.

[0091] The module supports data recording and storage. After each injection, the user can manually or automatically enter the specific injection site, drug dosage, injection device used, and possible adverse reactions such as subcutaneous bleeding, pain, or nodule formation. The system stores this data in a local or cloud database and analyzes it using data processing algorithms to identify long-term trends. For example, if a patient repeatedly experiences fatty hyperplasia or nodules in a specific area, the module will prompt the user to avoid that area and recommend alternatives.

[0092] In terms of long-term monitoring, the module can generate historical data curves to show the distribution of patients' injection sites at different time periods and correlate them with drug absorption effects. For diabetic patients, the system can synchronize data from blood glucose monitoring devices to evaluate the impact of different injection sites on blood glucose control and optimize future injection site recommendations. In addition, users can access historical injection data through the interface to understand the injection frequency of a specific area, ensuring that the rotation injection strategy is effectively implemented and avoiding fat hyperplasia or tissue damage caused by long-term use of the same area.

[0093] The module's interactive modes include touchscreen operation, voice control, and remote monitoring capabilities. For handheld devices, the system uses an intuitive touch interface, allowing patients to easily browse data, view injection recommendations, and record information. For smart wearable devices, the system can receive user feedback through voice commands and provide brief injection instructions in voice. In addition, the module can be connected to a telemedicine system, allowing medical staff to access the patient's injection records and adjust treatment plans or provide personalized recommendations when necessary.

[0094] To enhance ease of use, the module supports multiple data transfer methods, including local storage, Bluetooth synchronization, Wi-Fi connectivity, and cloud storage. For home users, the system can be configured for local storage, minimizing the risk of data transmission while ensuring privacy. For patients requiring telemedicine support, the module can upload data to an encrypted cloud server for remote analysis and treatment adjustments.

[0095] Overall, the module's design not only enhances the intelligence of subcutaneous injections, enabling patients to receive personalized injection recommendations, but also provides long-term monitoring and dynamic optimization capabilities to ensure injection safety and stable drug absorption. Through an intuitive user interface, interactive operation, adaptive data analysis, and remote monitoring capabilities, the module effectively supports long-term treatment management for patients in both home and clinical settings.

[0096] Furthermore, the user interaction module includes a visual feedback interface unit, an injection feedback collection unit and an individual behavior recognition unit, wherein the visual feedback interface unit is configured to spatially mark the personalized injection site recommendation plan in the form of a body surface topology map in the display interface, and use color coding to distinguish injectable areas, contraindicated areas and priority recommended areas; the injection feedback collection unit is configured to guide the user to enter the injection location, drug type and tissue reaction within 48 hours after the injection is completed, including the area of redness and swelling, pain level and tissue tactile changes, and automatically associate and store the feedback information with the recommended injection point; the individual behavior recognition unit identifies the patient's compliance, tendency and injection habits in actual use based on the user's long-term injection data and feedback pattern, and adjusts the weight parameters of future recommendation strategies to achieve cross-user personalized modeling and adaptive optimization.

[0097] In this embodiment, the user interaction module is designed to have multi-level and multi-dimensional interactive functions, specifically including a visual feedback interface unit, an injection feedback acquisition unit and an individual behavior recognition unit, and is highly integrated with the data flow, model calculation and result presentation logic of the entire injection site detection and recommendation system to form a complete closed-loop interactive system, which not only improves the user experience, but also enhances the system's adaptability to individual differences.

[0098] Among them, the visual feedback interface unit is used to present personalized injection site recommendation results to the user in an intuitive manner. After the system analyzes the subcutaneous tissue characteristic data of the current patient and generates a recommendation plan, the recommendation results are displayed on the device screen in the form of a surface topology map. For example, a simplified outline of the human trunk or upper limb is used as the background, all allowed injection areas are depicted in gray, and color coding is used to distinguish the injection risk levels of various areas. For example, green indicates the preferred injection point, yellow indicates the area that can be used but is recommended to be rotated, and red indicates high-risk areas where injection should be avoided, such as areas with dense blood vessels, previous induration areas, or areas with abnormal tissue reactions. Each point or area in the interface not only visually marks its risk level, but also provides a pop-up window after clicking, showing the detailed injection history, tissue status assessment results, model score and other information of the point, which is convenient for users to refer to and make decisions.

[0099] The injection feedback collection unit is a key component in achieving the system's long-term adaptability. It is configured to guide the user in recording specific information about the injection after the injection is completed. The user can select the injection location through touch or choose to mark the coordinates of the current injection point using camera-assisted positioning. The system prompts the user to enter or select the drug name and dosage, and within 48 hours after the injection, reminds the user to fill in the injection reaction information, including but not limited to the area of redness and swelling (for example, in millimeters or with a graphic), the pain level (using a subjective rating of 0 to 10), and changes in tissue tactile sensation (such as whether the touch pressure feels hard or sensitive). This module supports multiple input methods, such as sliders, selectable options, and voice-to-text, to adapt to the operating habits of users of different age groups. After receiving feedback, the system automatically binds this feedback data to the recommended injection point, forming a complete set of injection record events, which is stored in the patient's personal injection database for subsequent analysis.

[0100] The individual behavior recognition unit is based on the injection behavior and feedback data recorded by the system over a long period of time, from which it models the user's usage preferences, compliance, and injection tendencies. This unit determines the user's acceptance of the system's recommendations by counting the offset distance between the user's actual selection point and the recommended point in several past recommendation plans. It can also analyze whether the user's injection frequency in a specific area is too high, and whether there is an increase in adverse reactions due to repeated injections in a certain tissue area. In addition, the system can also identify whether the user is more sensitive to pain, whether he prefers to stay away from vascular areas, or whether he has uneven rotation usage behavior. Based on the above behavioral modeling results, the system will automatically adjust the weight coefficients in the subsequent injection site recommendation algorithm, such as reducing the recommendation priority of areas that the user has repeatedly ignored, or increasing the dynamic buffering of risk boundaries in the user's preferred areas, so as to better conform to the user's operating habits and tissue state changes, and realize individualized adjustment of the recommendation strategy.

[0101] Through the coordinated operation of these three functions, the user interaction module not only serves as an information output terminal to present results, but also becomes an important data source and policy control entry point for the adaptive evolution of the system. This module is particularly suitable for patients with chronic diseases who require long-term regular injections, such as those with insulin-dependent diabetes or users of long-term biologic injections. It not only improves the convenience of injection management but also effectively avoids complications caused by improper injection site management. Relying on modern touch screen devices and image recognition technology, this module can be implemented on common smart terminals, showing strong engineering feasibility and user-friendliness.

[0102] Furthermore, the user interaction module is configured to generate a user-specific injection management cycle, and dynamically lock the recommended injection site plan or provide risk upgrade prompts based on the injection results and complication reports input by the user. The user interaction module includes a cycle guidance unit, which is used to generate an injection rotation sequence and a minimum injection interval between different anatomical areas on the body surface based on the recommended areas identified by the system, and prompt the user to perform the injection as planned in a graphical calendar format; the complication perception unit is used to receive local erythema, persistent nodules, and infection signs actively input by the user or identified by the system, and upon detecting that a specific risk indicator exceeds a preset threshold, trigger the temporary closure of the corresponding injection area, risk warning, and doctor intervention prompt functions, thereby realizing a real-time interactive response mechanism based on the dynamic changes of complications.

[0103] In this embodiment, the user interaction module not only presents injection recommendations but also dynamically generates an injection management cycle based on individual status and feedback. It also continuously monitors and dynamically adjusts injection risks to achieve long-term, safe, efficient, and personalized injection plan management. By establishing a patient-specific injection management cycle, supplemented by a cycle guidance mechanism and complication awareness mechanism, the system forms a closed-loop interactive injection planning and risk management process.

[0104] Specifically, after identifying the current injectable area, the system will divide the injection area into zones based on the anatomical features of the body surface, for example, into multiple sub-areas such as the left side of the abdomen, the right side of the abdomen, the left upper arm, the right upper arm, the left thigh, and the right thigh. Within each area, the system automatically generates an injection rotation sequence and the corresponding minimum injection interval based on the preferred injection point and local risk level determined by the scoring function. This interval can be calculated based on factors such as the patient's tissue recovery time, the drug metabolism cycle, and the intensity of tissue reaction, and is usually adjusted dynamically within the range of 48 to 96 hours. The system presents the recommended daily injection areas and alternative locations to the user in a color-coded manner through a calendar-style visual interface. For example, the currently recommended area is marked green, the area that was previously injected but has not yet recovered is marked gray, and the non-recommended area is marked red. Icons are used to remind users to follow the rotation rules to avoid repeated use of local tissues.

[0105] After the injection is completed, the system guides the user to submit injection feedback, including the injection site, time, drug information, and subjective feelings and objective observations within 48 hours after the injection, such as whether erythema, local nodules, skin temperature changes, or exudation occur. The system can also automatically record changes in local skin conditions through image capture, such as capturing the range and color distribution of redness and swelling in the injection area, and quantifying the erythema area or nodule shadow characteristics through image recognition algorithms. In addition, if the system is integrated with a bioimpedance sensing function, it can also obtain changes in the conductivity of the local tissue at that point to determine the degree of inflammation or tissue edema.

[0106] All of the above information will be aggregated to the complication perception unit, which will conduct a real-time assessment of the risk status of the current injection point based on the set multi-dimensional threshold model. For example, if two consecutive nodule feedbacks appear in a certain area, or the area of erythema exceeds the preset area threshold (such as a diameter greater than 25 mm), the system will automatically temporarily block the injection area and prohibit the continued recommendation of this area in subsequent injection management cycles. At the same time, the system can issue a risk warning prompt to the user and pop up a prompt window suggesting that the user contact a doctor; if the system is connected to a telemedicine platform or a patient management platform, the risk data can be automatically uploaded, and an event notification record can be established on the doctor's side to form a closed-loop mechanism for treatment intervention.

[0107] By integrating this cycle guidance with the complication perception mechanism, the system dynamically updates injection recommendation strategies based on individualized physiological responses and behavioral feedback, evolving from static recommendations to dynamic, adaptive management. Without incurring additional operational burdens, users can access a continuously optimized, risk-controlled, long-term injection management plan. This approach is particularly suitable for chronic disease treatment scenarios requiring daily or alternate-day injections, such as insulin therapy, anticoagulant injections, or the long-term administration of biologics.

[0108] Furthermore, the user interaction module includes a body surface guidance calibration unit, which is configured to, after generating a personalized injection site recommendation plan, calibrate the spatial matching relationship between the user's current position and the recommended injection point in combination with the system positioning component or image re-recognition function, and prompt the user to complete precise injection positioning through a multi-modal method of screen guidance, vibration feedback, and image overlay; the body surface guidance calibration unit is also configured to, after the user manually confirms or touches the selected injection point, capture the injection site image through the camera, perform feature alignment with the original recommendation layer, and determine whether the actual injection point deviates from the recommended area. If the deviation exceeds the system set tolerance, a re-positioning prompt and a deviation risk prompt are triggered.

[0109] In this embodiment, the user interaction module is configured to include a body surface guidance calibration unit, which is used to help users accurately locate and execute the system-recommended injection points. This unit avoids injection point selection errors caused by visual errors, body position changes, or subjective judgment errors in actual operation, thereby improving the execution accuracy of the recommended plan and the consistency of treatment effect. By integrating spatial positioning, image recognition, and multimodal feedback technologies, the body surface guidance calibration unit establishes a real-time spatial mapping relationship between the user's current body surface state and the system-recommended points, and dynamically corrects deviations during the interaction process, achieving precise alignment between the user and the recommended area.

[0110] In practice, after the system generates a personalized injection site recommendation, the target injection point is graphically spatially annotated in the user interface. To ensure that the user can accurately locate the recommended point on their body surface, the system first identifies the current device's orientation and relative position based on built-in positioning components (such as the IMU inertial measurement unit, gyroscope, and camera). It then establishes a registration model based on the screen display orientation and the user's body surface coordinates. If the device has image recognition capabilities, the system can also extract reference structures of the human body region through key surface points (such as the navel, shoulders, and joint lines) in real-time camera images, further narrowing the positioning error range of the target injection area.

[0111] When prompting users to perform positioning operations, the system uses multimodal interaction to enhance guidance. For example, a simplified map of the human body area is displayed on the screen in real time, and the user is gradually guided to move the device to the correct position through dynamic indicator points, arrow paths, or halo animations. At the same time, the device can also incorporate a haptic feedback module to emit a vibration prompt when approaching the recommended injection area, or use voice prompts to guide the user to adjust the angle or direction until the positioning deviation is within the set tolerance range.

[0112] Once the user confirms the selected injection point by clicking on the screen or by voice, the system will start the calibration process, using the front or rear camera to capture an image of the injection area and align the image features with the recommendation layer originally generated by the system. This alignment process uses local texture matching, shape boundary comparison, or marker point positioning to analyze the spatial offset between the user-selected injection point and the recommended point. If the offset is within the set tolerance range, for example, less than 5 mm, the system confirms that the positioning is successful; if the offset exceeds the threshold, a re-positioning prompt is triggered, and a warning box is displayed in the interface to prompt the user that there is an execution error in the current position, and it is recommended to realign or use the visual overlay mode for further calibration.

[0113] Furthermore, the system can provide intelligent prompts based on the type of deviation. For example, if a user mistakenly selects a high-risk area, the system will trigger both a "risk warning" and a "recalibration suggestion," showing the user the corresponding tissue characteristics and potential complication risks, to help them make safer operational decisions. All positioning and calibration actions are recorded by the system and incorporated into long-term user behavior analysis data to improve the behavioral adaptability of the system's subsequent recommendation strategies.

[0114] This implementation significantly improves the consistency of the system's recommendations during real-world procedures, making it particularly suitable for users who need to manipulate the injection site themselves, such as the elderly, those with poor vision, or those with mobility issues. This technology leverages existing camera recognition technology, human posture detection, and mobile device inertial positioning capabilities, combined with multimodal human-computer interaction, to form a complete spatial calibration loop, ensuring accuracy, safety, and user compliance during subcutaneous injections.

[0115] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A subcutaneous injection site detection and recommendation system based on intelligent algorithms, characterized in that: include: a subcutaneous tissue detection module configured to detect the subcutaneous blood vessel distribution, fat layer thickness, and induration status of a target site through infrared imaging and ultrasonic technology, and generate corresponding subcutaneous tissue characteristic data; a data processing module, in communication with the subcutaneous tissue detection module, configured to receive the subcutaneous tissue characteristic data and identify vascular dense areas, indurations, and areas where fat hyperplasia has occurred through an image recognition algorithm; An injection site recommendation module analyzes the recognition results of the data processing module based on a machine learning algorithm and combines the patient's personalized injection history data to generate personalized injection site recommendations that avoid areas with dense blood vessels, nodules, and areas where fat hyperplasia has occurred, thereby reducing the risk of complications related to subcutaneous bleeding, subcutaneous nodules, and subcutaneous fat hyperplasia, while optimizing drug absorption efficiency. The user interaction module is configured to present the personalized injection site recommendation to the user and record the site and effect data of each injection to support long-term monitoring and treatment adjustment.

2. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 1, characterized in that: The subcutaneous tissue detection module includes an infrared imaging unit and an ultrasonic imaging unit, wherein the infrared imaging unit is configured to collect thermal images of the target part under two or more different infrared bands, and highlight the direction of blood vessels and thermal abnormality areas through image superposition and contrast enhancement processing; the ultrasonic imaging unit is configured to emit ultrasonic signals of different frequencies in sequence, and determine the thickness of the fat layer and the boundary of the abnormal high-density area based on the change in echo intensity at each frequency; the images collaboratively acquired by the infrared imaging unit and the ultrasonic imaging unit are processed through spatial coordinate alignment to form subcutaneous tissue feature data including blood vessel distribution, fat layer thickness and abnormal tissue boundary information.

3. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 2, characterized in that: The data processing module includes an image fusion receiving unit, a fusion image deconstruction unit and a feature data structure generation unit, wherein: The image fusion receiving unit is configured to receive a fusion image output by the subcutaneous tissue detection module after spatial coordinate alignment processing, wherein the fusion image includes the vascular thermal response distribution, fat layer thickness information and abnormal tissue boundary in the target area; The fused image deconstruction unit is configured to deconstruct image signals corresponding to different tissue types in the fused image by a layered analysis method, including extracting a thermal gradient trend layer associated with blood vessels, an echo amplitude distribution layer associated with fat layers, and a high-density boundary marker layer associated with induration in the fused image; The feature data structure generating unit is configured to generate first subcutaneous tissue feature data in a unified format based on the multi-layer image information extracted by the fused image deconstruction unit, including the spatial distribution, boundary range and tissue attribute labels of various risk areas in a unified coordinate system.

4. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 3, characterized in that: The injection site recommendation module constructs a scoring function based on the support vector machine model to quantitatively score the injection suitability of each candidate injection point. The scoring function is defined as the following formula 1: in, is the number of historical sample points used to train the support vector machine model, each of which has a labeled feature vector and spatial coordinates; is a symbolic function; represents the injection suitability score; Indicates candidate injection sites The corresponding eigenvector; For the The support vector weights corresponding to the training samples are obtained through support vector machine model training; For the training set The feature vector of each sample; is the kernel width parameter, which is used to adjust the mapping scale of the feature space; is the model bias term; is the position weight function; wherein the feature vector is expressed by the following formula 2: in, Candidate injection sites The vascular density coefficient at a point is calculated based on the number of vascular thermal responses detected per unit area in a local window centered at the point in the fused image; Candidate injection sites The vertical thickness of the fat layer was derived from the depth measurement of the fat tissue layer in ultrasound images; Candidate injection sites The proportion of abnormal tissue in the surrounding area is defined as The percentage of the area identified as induration, scar or fat hyperplasia in the image to the entire neighborhood area within the center; Candidate injection sites The historical injection frequency is provided by the user injection record system and is defined as the ratio of the number of times the point is used in the user's historical injection distribution to the total number of injections; Indicates candidate injection points The corresponding tissue reaction level label is obtained by evaluating the user's feedback data within 48 hours after the injection at the same point. The feedback data includes the area of redness and swelling, subjective pain score, and the amplitude of the change in local bioelectrical impedance; Position weight function Calculated by the following formula 3: in, Candidate injection sites With training sample points Euclidean distance in body surface space; is the structural risk adjustment factor; The candidate points with positive injection suitability scores are determined as injectable areas.

5. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 2, characterized in that: The infrared imaging unit is configured to acquire time-series thermal images in a plurality of continuous infrared bands, and extract dynamic blood flow characteristics by analyzing the temperature gradient change rate, so as to distinguish the direction of veins and arteries.

6. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 2, characterized in that: The ultrasonic imaging unit is configured to construct a tissue elasticity map based on the echo phase difference, which is used to assist in identifying potential nodule areas that have not formed obvious density differences in the early stage, thereby improving the early detection capability of abnormal tissue and enhancing the accuracy of risk area identification.

7. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 3, characterized in that: The fused image deconstruction unit is configured to extract composite risk areas based on the pixel-level spatial overlap relationship between the layers, and assign a high-risk weight label to the intersection area of the vascular thermal response and the abnormal density boundary; The characteristic data structure generation unit is configured to encode various risk areas in layers according to risk levels, and establish time-series difference data between consecutive frame images for analyzing the dynamic change trend of local tissue status, thereby improving the timeliness and accuracy of injection site risk assessment.

8. The subcutaneous injection site detection and recommendation system based on intelligent algorithm according to claim 1, characterized in that: The user interaction module includes a visual feedback interface unit, an injection feedback collection unit and an individual behavior recognition unit, wherein the visual feedback interface unit is configured to spatially mark the personalized injection site recommendation plan in the form of a body surface topology map in the display interface, and use color coding to distinguish injectable areas, contraindicated areas and priority recommended areas; the injection feedback collection unit is configured to guide the user to enter the injection location, drug type and tissue reaction within 48 hours after the injection is completed, including the area of redness and swelling, pain level and tissue tactile changes, and automatically associate and store the feedback information with the recommended injection point; the individual behavior recognition unit identifies the patient's compliance, tendency and injection habits in actual use based on the user's long-term injection data and feedback pattern, and adjusts the weight parameters of future recommendation strategies to achieve cross-user personalized modeling and adaptive optimization.

9. The intelligent algorithm-based subcutaneous injection site detection and recommendation system according to claim 1, characterized in that: The user interaction module is configured to generate a user-specific injection management cycle, and dynamically lock the recommended injection site plan or provide risk upgrade prompts based on the injection results and complication reports input by the user. The user interaction module includes a cycle guidance unit, which is used to generate an injection rotation sequence and a minimum injection interval between different anatomical areas on the body surface based on the recommended areas identified by the system, and prompt the user to perform the injection as planned in a graphical calendar format; the complication perception unit is used to receive local erythema, persistent nodules, and infection signs actively input by the user or identified by the system, and upon detecting that a specific risk indicator exceeds a preset threshold, trigger the temporary closure of the corresponding injection area, risk warning, and doctor intervention prompt functions, thereby realizing a real-time interactive response mechanism based on the dynamic changes of complications.

10. The subcutaneous injection site detection and recommendation system based on intelligent algorithm according to claim 1, characterized in that: The user interaction module includes a body surface guidance calibration unit, which is configured to, after generating a personalized injection site recommendation plan, calibrate the spatial matching relationship between the user's current position and the recommended injection point in combination with the system positioning component or image re-recognition function, and prompt the user to complete precise injection positioning through a multi-modal method of screen guidance, vibration feedback, and image overlay; the body surface guidance calibration unit is also configured to, after the user manually confirms or touches the selected injection point, capture the injection site image through the camera, perform feature alignment with the original recommendation layer, and determine whether the actual injection point deviates from the recommended area. If the deviation exceeds the system set tolerance, a re-positioning prompt and a deviation risk prompt are triggered.

Images