A wound healing status detection and healing time estimation device and method

Abstract

This invention discloses a device and method for detecting wound healing status and predicting healing time. The device includes a flexible substrate on which a photoelectric detection area and a light source area surrounding the photoelectric detection area are formed; a photoelectric detection array disposed within the photoelectric detection area, the photoelectric detection array including at least one organic photodetector; multiple light sources disposed within the light source area and spaced circumferentially thereon, each light source having its emitting surface facing the photoelectric detection area, for sequentially irradiating wound tissue located above the device with multiple preset wavelengths; a positioning area disposed on the upper surface of the flexible substrate, the planar projection of the positioning area on the flexible substrate at least partially overlapping the photoelectric detection area; at least one pressure sensor disposed within the positioning area; an angle sensor disposed on the flexible substrate and / or a circuit board below it; and a processing unit including a microprocessor integrating an analog-to-digital converter, a storage unit, and an embedded computing unit. This invention also discloses a related prediction method.

Description

Technical Field

[0001] This application relates to medical and health monitoring technology, specifically, to a device and method for detecting wound healing status and predicting healing time. Background Technology

[0002] With the increasing demand for precision medicine and home-based health monitoring, real-time assessment of wound healing status has gradually become a crucial aspect of clinical nursing, chronic disease management, and postoperative rehabilitation. This is especially true for chronic or complex wounds such as diabetic foot ulcers, pressure ulcers, and postoperative incisions. Failure to promptly and accurately monitor their healing status and trends can easily lead to serious consequences such as infection, delayed healing, or even amputation. Therefore, objective, quantitative, and continuous monitoring of the wound healing process has become a common need in both clinical and home settings.

[0003] Existing wound healing assessment techniques can be broadly categorized into the following three types: The first category is indirect assessment methods based on biochemical indicators. A typical approach involves detecting systemic metabolic or inflammation-related indicators such as blood lactate and inflammatory factors to indirectly infer whether the wound is ischemic, infected, or experiencing a systemic stress response. This type of method has some application in fields such as intensive care and sports medicine, but it has the following limitations: First, the tests largely rely on invasive blood collection and reagent testing, causing additional trauma to the patient and making it unsuitable for frequent, repeated use. Second, the obtained indicators mainly reflect the overall metabolic or inflammatory state of the body, and their correlation with the healing degree of a specific wound is relatively indirect, making it difficult to provide precise quantitative assessments of the healing depth, granulation quality, and epithelialization degree of a single wound. Third, the cost of reagents and consumables, as well as the testing process, are relatively complex, making long-term, continuous monitoring in home or community settings unsuitable.

[0004] The second category is medical imaging detection technology. Typical methods include using high-frequency ultrasound imaging to detect granulation tissue thickness and using infrared thermal imaging to monitor local temperature changes. These methods can provide certain objective images or signal parameters, offering better objectivity compared to purely subjective assessments. However, these devices are usually large and expensive, require a certain level of technical expertise from operators, and are mainly deployed in medical institutions, making them unsuitable for high-frequency, long-term monitoring of patients in their home environments. Furthermore, methods such as infrared thermal imaging are easily affected by external factors such as ambient temperature and measurement distance, resulting in limited accuracy and stability in home settings with significant room temperature fluctuations, making continuous tracking of individual wounds difficult.

[0005] The third category is optical reflectance detection technology. This type of technology is based on the correlation between the optical properties of biological tissues and reflectance. It illuminates wound tissue with a specific wavelength light source, and then a photodetector receives the reflected light signal to infer the tissue state. Existing solutions mostly adopt a structure of "single-wavelength LED / laser light source + silicon-based photodetector". Although it has advantages in miniaturization and portability to a certain extent, and is considered a potential direction suitable for home or wearable applications, it is limited by hardware layout and algorithm design. This type of solution generally has the following problems: On the one hand, it is difficult to keep the contact posture of the finger or limb with the detector head consistent during detection. Small changes in pressure and angle can significantly change the path of incident and reflected light, resulting in large fluctuations in reflectance readings. It lacks an effective pressure / angle calibration and compensation mechanism and has insufficient anti-interference ability. On the other hand, most devices only output raw reflectance data at a single wavelength or a few wavelengths. They are not deeply integrated with artificial intelligence algorithms such as deep learning, making it difficult to comprehensively consider multi-wavelength reflectance characteristics and individual physiological parameters (such as age, blood sugar level, etc.) to give a quantitative result of wound healing level, and even more so, it cannot reliably predict the remaining healing time. From a device perspective, using organic light-emitting diodes (OLEDs) as the light source and organic photodetectors (OPDs) as the receiver offers advantages over traditional inorganic LED + silicon photodiode solutions. These advantages include customizable wavelengths, ease of fabrication into flexible and large-area arrays, and high-integration packaging on the same flexible substrate. This promises to enable multi-wavelength, close-range, skin-adherent optical reflection detection while significantly reducing device size and thickness. However, current technologies show limited integration of OLED light sources and OPD arrays in wound detection. There is a lack of flexible modules with an integrated layout of ring-shaped multi-wavelength emission and central array detection, specifically designed for small wounds such as those on fingers. Furthermore, there is a lack of supporting attitude correction models and complete solutions deeply integrated with AI prediction algorithms, hindering the full realization of the potential advantages of OLEDs and OPDs in multi-wavelength detection, flexible adhesion, and system integration.

[0006] In summary, existing technologies have the following main shortcomings in wound healing assessment: 1. It relies on invasive blood collection and reagents, only reflects the overall metabolic and inflammatory state of the body and is difficult to quantify the degree of healing of individual wounds in detail. In addition, the testing process and consumable costs are high, making it unsuitable for high-frequency, long-term continuous monitoring in home settings. 2. Medical imaging equipment is expensive, bulky, and has high requirements for usage scenarios and operators, making it difficult to meet the daily monitoring and continuous tracking needs in home settings; 3. Existing optical reflection detection devices have not been systematically optimized for uneven wound surfaces and changes in contact posture, making them susceptible to interference from pressure intensity and angle, resulting in large errors in reflectivity detection; 4. It lacks organic integration with artificial intelligence models, cannot convert optical signals into interpretable healing levels, and cannot achieve intelligent prediction of healing time based on individual characteristics.

[0007] Therefore, there is a need for a new technology solution for wound healing status detection and healing time estimation that is structurally anti-interference, algorithmically implements reflectivity calibration and intelligent prediction of healing time, and is also miniaturized and portable, in order to make up for the shortcomings of existing technologies in accurate assessment and continuous home monitoring. Summary of the Invention

[0008] To overcome the above-mentioned shortcomings of the prior art, the present invention provides a device and method for detecting wound healing status and predicting healing time, which solves the problem of reflectivity reading fluctuation caused by inconsistent contact posture in the existing optical reflectivity detection technology, and realizes stable and high-precision detection of wound tissue.

[0009] To achieve the above objectives, the present invention adopts the following technical solution.

[0010] A device for detecting wound healing status and predicting healing time based on optical reflectivity, comprising: A flexible substrate on which a photoelectric detection area and a light source area surrounding the photoelectric detection area are formed; A photoelectric detection array is set in the photoelectric detection area, and the photoelectric detection array includes at least one organic photodetector; Multiple light sources are set up in the light source area and arranged at intervals along its circumference. The light-emitting surface of each light source faces the photoelectric detection area and is used to irradiate the wound tissue located above the device in sequence with multiple preset wavelengths. A positioning area is set on the upper surface of a flexible substrate, and the planar projection of the positioning area on the flexible substrate at least partially overlaps with the photoelectric detection area, for guiding the wound area of ​​the finger or limb to align with the photoelectric detection array. At least one pressure sensor is disposed in the positioning area for detecting the contact pressure between the finger and the device; An angle sensor, mounted on a flexible substrate and / or a circuit board beneath it, is used to detect the spatial orientation of the device or finger pressing direction. The processing unit includes a microprocessor that integrates an analog-to-digital converter, a storage unit, and an embedded computing unit. The processing unit is electrically connected to the photoelectric detection array, the pressure sensor, the angle sensor, and the light source, respectively.

[0011] This application uses a pressure sensor and an angle sensor to form an attitude sensor that measures in the positioning area. By introducing an attitude sensor and an intelligent processing unit, it effectively solves the problems of optical reflection detection being susceptible to attitude interference and lacking intelligent prediction. At the same time, through flexible and integrated device design, it improves the portability and applicability of the device, providing an innovative technical solution for wound healing status detection and healing time prediction in home settings.

[0012] Furthermore, this application proposes that the flexible substrate is a polyimide substrate, covered with a medical-grade elastic encapsulation layer, the encapsulation layer forming a positioning area on the upper surface for direct contact with the skin and providing flexible adhesion.

[0013] By employing polyimide as a flexible substrate and covering it with a medical-grade elastic encapsulation layer, a positioning area is formed on the upper surface of the flexible substrate. This allows the entire device to maintain its core optical detection function while significantly enhancing its adhesion to human skin, comfort, and biocompatibility. This ensures that the device can provide more stable and accurate detection data in practical applications, laying a solid foundation for subsequent processing units to assess wound healing status and predict healing time.

[0014] Furthermore, this application proposes that the light source has a preset tilt angle relative to the flexible substrate to form annular multi-wavelength oblique incidence illumination. This allows the light to enter the wound tissue along a more optimized path and be scattered, increasing the probability and stability of the photoelectric detection array capturing effective reflected light. The annular arrangement, combined with multi-wavelength characteristics, ensures uniform and comprehensive illumination of the wound area and provides rich spectral information. The combination of oblique incidence illumination and annular multi-directional incidence effectively reduces interference caused by changes in the user's finger or limb pressure or angle of contact on reflectivity readings, making the collected reflected light signals more stable and accurate.

[0015] Furthermore, this application also proposes that the shape of the photoelectric detection area corresponding to the flexible substrate is a groove disposed at the center of the flexible substrate, and the shape of the light source area corresponding to the flexible substrate is an annular groove disposed around the outside of the photoelectric detection area.

[0016] By designing the photodetector area on the flexible substrate as a central groove, a stable and precise physical positioning area is provided for the photodetector array, ensuring that it is always at the geometric center of the detection area. When the user places wound tissue on the device, the central groove guides the wound area to align with the photodetector array, while the annular groove ensures that multiple light sources can illuminate the wound tissue at a consistent, circumferential angle. This effectively avoids the problem of unstable incident and reflected light paths caused by changes in contact pressure or angle in traditional planar designs.

[0017] Furthermore, this application proposes that the light sources are uniformly arranged circumferentially along the annular light source region, and the light sources include at least four sets of OLED light sources with emission wavelengths covering 450–950 nm. By uniformly arranging the light sources circumferentially along the annular light source region, it is ensured that light is uniformly irradiated onto the wound tissue located above the photodetector array from all directions, thereby avoiding reflectivity measurement deviations caused by uneven local illumination. Simultaneously, the inclusion of at least four sets of OLED light sources provides sufficient light source quantity and redundancy, enhancing the comprehensiveness and robustness of illumination. The emission wavelength coverage of 450–950 nm enables the device to acquire rich optical information of biological tissue in the visible to near-infrared bands.

[0018] This application also discloses a prediction method based on the above-mentioned wound healing status detection and healing time prediction device, including the following steps: Step A, Posture Guidance and Contact Detection: The user places the injured finger or other wound on the positioning area of ​​the device. The pressure sensor detects whether the contact pressure is within the preset range, and the angle sensor detects whether the pressing angle is within the preset range. When the pressure and angle meet the set conditions, a contact status compliance prompt is output, allowing the process to proceed to the optical detection stage. Otherwise, the user is prompted to adjust the pressing posture through an audio-visual signal. Step B: Multi-wavelength optical illumination and reflection signal acquisition; control the OLED light source of each wavelength to emit light in a preset order, and acquire the multi-wavelength reflected photoelectric signals output by the photoelectric detection array as well as the corresponding pressure and angle information; according to the signal processing and the pre-stored "pressure-angle-reflectivity" correction model, correct the original reflectivity obtained under different pressures and angles to the corrected reflectivity equivalent to the standard posture. Step C: Output of wound healing quantification level; The corrected multi-wavelength reflectance is used as input and fed into the wound quantification correlation model trained with clinical annotation data to output the corresponding wound healing level; The healing level is divided into multiple levels according to preset standards to characterize different healing stages from superficial repair, granulation growth, deep healing to severe damage. Step D: Healing time prediction and risk assessment; Based on historical test data and / or individual physiological characteristics, call the pre-stored healing time prediction model to output the remaining wound healing time and / or healing delay risk warning; Step E, Result Display and Data Storage: The current wound healing level, estimated healing time, and risk warnings are visualized and output through the display module or external terminal. At the same time, information such as the multi-wavelength reflectivity, attitude parameters, and prediction results of this test are stored locally or in the cloud for subsequent continuous monitoring and incremental model optimization.

[0019] This invention establishes a complete method for wound healing status detection and healing time prediction by integrating a series of steps, including posture guidance and contact detection, multi-wavelength optical illumination and reflection signal acquisition, quantitative output of wound healing level, healing time prediction and risk assessment, and result display and data storage. The method utilizes hardware components integrated on a flexible substrate, such as a photoelectric detection array, OLED light source, pressure sensor, and angle sensor. Through a correction model and AI prediction model embedded in the processing unit, it achieves full automation from raw signal acquisition to intelligent prediction. This application effectively solves the problems of large fluctuations in reflectivity data and lack of effective calibration mechanisms caused by inconsistent operating postures in traditional wound healing assessments, significantly improving the accuracy and reliability of detection data. Simultaneously, by introducing a wound quantification correlation model and a healing time prediction model trained based on clinically labeled data, it achieves objective quantitative assessment of wound healing status and personalized intelligent prediction of healing time, overcoming the shortcomings of existing technologies in intelligent prediction.

[0020] Furthermore, this application also proposes that, in step D, it be determined whether historical detection data for the wound exists in the storage module: When there is no historical data, the AI ​​processing module calls the single prediction mode based on the current healing level, wound type and individual physiological characteristics parameters, and outputs the estimated wound healing cycle or remaining healing days through a deep learning model. When historical data exists, the AI ​​processing module takes the reflectivity sequence and healing level sequence obtained from multiple detections as time series input, calls the trend prediction mode, uses a deep learning model to learn the reflectivity change trend and healing speed, outputs a more refined prediction result of the remaining healing time, and combines the model judgment to give early warning information such as healing delay and infection risk.

[0021] This method intelligently selects between a single-test prediction mode and a trend prediction mode based on the availability of historical wound monitoring data. When historical data is unavailable, the system provides a preliminary and reasonable healing time estimate based on the current healing level, wound type, and individual physiological parameters, addressing the issue of ineffective healing time prediction during initial testing or when continuous monitoring data is lacking. When historical data is available, the system fully utilizes reflectivity and healing level sequences obtained from multiple tests, employing a deep learning model to learn healing trends and speeds. This results in a more refined and accurate prediction of remaining healing time, along with timely warnings of healing delays and infection risks. This case-by-case approach significantly improves the accuracy and personalization of healing time prediction, enabling users to receive healing management recommendations more tailored to their specific wound conditions, effectively guiding wound care and reducing the risk of complications.

[0022] Furthermore, this application proposes the following specific steps for constructing the "pressure-angle-reflectivity" correction model: 1) Select standard samples with smooth surfaces and stable reflectivity, such as polished quartz sheets, as calibration objects and fix them in the detection area of ​​the device; 2) Within the preset pressure and angle range, design several pressure-angle combination points. At each combination point, apply the corresponding pressure and angle through the device, measure the original reflectivity Roriginal of multiple wavelengths multiple times, take the average value, and record the corresponding pressure P and angle θ to form a calibration dataset. 3) The reference reflectance R is measured under standard pressure and standard angle conditions. A simplified correction formula is obtained by fitting the deviations of each combination point through polynomial regression, for example: Rcorrected = Roriginal × (0.98 + 0.01P - 0.002θ) Where P is the pressure applied and θ is the pressure angle; 4) The error between the corrected reflectance and the reference reflectance is evaluated by cross-validation or partitioning the validation set. When the error meets the preset index, the corrected model parameters are stored in the microprocessor. In actual testing, the microprocessor calculates P and θ based on the real-time output of the pressure sensor and angle sensor, and calls the above correction formula to compensate for the original reflectivity, thus obtaining the target reflectivity R correction after attitude correction.

[0023] By providing high-precision attitude-corrected reflectivity data, the accuracy and reliability of subsequent wound healing quantification level output and healing time prediction are improved, enabling the entire detection system to obtain stable and reliable evaluation results even under non-professional user operation, thereby fully leveraging the advantages of continuous monitoring in home or grassroots environments.

[0024] Furthermore, this application also proposes that, in step C, the wound healing status L is marked as grade 1–10: grade 1–2: epidermal repair period; grade 3–5: granulation tissue growth period; grade 6–8: deep healing period; grade 9–10: severe injury or significantly delayed healing; For the data samples, the corrected reflectance was measured at multiple wavelengths, and the corresponding clinical healing levels were recorded. When establishing the wound healing level-multi-wavelength reflectance correlation model, the attitude-corrected multi-wavelength reflectance was used as a feature vector: Rcorrection = [Rcorrection(λ1), Rcorrection(λ2), ..., Rcorrection(λn)] The model takes the ratio / difference features derived from the wound healing level as input and the wound healing level L labeled by clinicians as supervision label. It is trained using supervised learning algorithms such as support vector machine, multi-class logistic regression, and neural network to obtain a nonlinear mapping model for realizing L=f(R correction). During online detection, the current multi-wavelength corrected reflectance is input into the model, and the corresponding wound healing level is output.

[0025] The implementation of this solution enables the device to obtain high-precision corrected reflectance R-correction equivalent to that under standard posture, regardless of the user or their operating habits. This improves the accuracy and stability of wound healing status detection, reduces the skill requirements for users, and allows non-professional users to perform reliable self-monitoring in a home environment. By providing accurate and consistent posture-corrected reflectance data, this solution provides high-quality input for the aforementioned wound healing quantification level output and healing time prediction methods, thereby significantly improving the objectivity of healing level assessment and the reliability of healing time prediction. This avoids misjudgments or delayed warnings caused by measurement errors, ultimately enhancing the clinical value and user experience of the entire wound healing management system.

[0026] Furthermore, this application proposes that the healing time prediction model is a deep learning model that integrates LSTM and CNN, and its structure includes: LSTM layer: The input dimension is the time series of multi-wavelength reflectance and / or healing level, used to learn the time-dependent features of reflectance changes during the healing process; CNN layer: Convolves the reflectance of multiple wavelengths in a two-dimensional space of "wavelength-time" to extract spatial correlation features between multiple wavelengths; Feature fusion layer: The temporal features output by LSTM and the spatial features output by CNN are concatenated and input into the fully connected layer along with the patient's physiological features; Output layer: The output is a regression value of the remaining days of healing or the time to complete healing, and can also output the classification results of the risk of delayed healing.

[0027] This fusion architecture of LSTM and CNN overcomes the limitations of single models when processing complex biomedical data. LSTM captures the dynamic process of wound healing, while CNN reveals the biophysical significance behind multi-wavelength reflectivity. The combination of these two technologies, supplemented by individual patient physiological characteristics, enables the model to understand the wound healing status from multiple dimensions and levels, thereby providing more accurate and personalized healing time predictions and risk assessments, significantly improving the accuracy and reliability of wound healing predictions.

[0028] The beneficial effects of this application are: 1. Improved anti-interference ability; 2. Strong ability to quantify healing status; 3. High integration and good portability; 4. Accurate and intelligent prediction of healing time; 5. Good biocompatibility and comfort. Attached Figure Description

[0029] Figure 1 This is a top view of the structure of a wound healing detection device in one embodiment of the present invention.

[0030] Figure 2This is a cross-sectional view of the wound healing detection device structure in one embodiment of the present invention.

[0031] Figure 3 This is a flowchart illustrating the overall workflow of healing detection in one embodiment of the present invention.

[0032] Figure 4 The flowchart shows the construction process of the pressure-angle-reflectivity model in one embodiment of the present invention.

[0033] Figure 5 This is a flowchart illustrating the construction of a wound quantification level-reflectivity model in one embodiment of the present invention.

[0034] Figure 6 The flowchart below shows the construction process of a wound healing time prediction model in one embodiment of the present invention.

[0035] In the figure: flexible substrate 101, photoelectric detection area 102, light source area 103, photoelectric detection array 201, light source 202, positioning area 104, pressure sensor 301, angle sensor 302, processing unit 401, elastic encapsulation layer 405, display unit 501, buzzer 502. Detailed Implementation

[0036] Example 1, A device for detecting wound healing status and predicting healing time based on optical reflectivity, such as... Figure 1 and Figure 2 As shown, it includes: A flexible substrate 101 has a photoelectric detection area 102 and a light source area 103 arranged around the photoelectric detection area; The photoelectric detection array 201 is disposed in the photoelectric detection area 102, and the photoelectric detection array includes at least one organic photoelectric detector; Multiple light sources 202 are arranged at intervals along the circumference within the light source area 103, with the emitting surface of each light source facing the photoelectric detection area 102, and are used to sequentially irradiate the wound tissue located above the device with multiple preset wavelengths. The positioning area 104 is disposed on the upper surface of the flexible substrate 101. The planar projection of the positioning area on the flexible substrate at least partially overlaps with the photoelectric detection area 102, which is used to guide the wound area of ​​the finger or limb to align with the photoelectric detection array 201. At least one pressure sensor 301 is disposed in the positioning area 104 for detecting the contact pressure between the finger and the device; An angle sensor 302 disposed on the flexible substrate 101 and / or the circuit board below it is used to detect the spatial posture of the device or finger pressing direction. The processing unit 401 includes a microprocessor that integrates an analog-to-digital converter, a storage unit, and an embedded computing unit. The processing unit is electrically connected to the photoelectric detection array 201, the pressure sensor 301, the angle sensor 302, and the light source 202, respectively.

[0037] For ease of understanding, the following explains some key terms in this embodiment: The flexible substrate 101 is designed as a bendable and deformable structure to support various electronic components in the device. Its flexibility allows it to conform to the surface of human skin and adapt to the curvature of different areas, thereby improving the stability and comfort of the testing process.

[0038] The photoelectric detection area 102 is a region specifically designated on the flexible substrate 101 for receiving reflected light signals. This area is typically located at the center of the device to facilitate alignment with the wound tissue to be detected.

[0039] The light source area 103, which surrounds the photoelectric detection area 102, is used to install multiple light sources. The layout of this area is designed to provide uniform illumination, ensuring that the wound tissue is adequately irradiated.

[0040] The photoelectric detection array 201, which consists of multiple photoelectric detectors, is integrated within the photoelectric detection area 102. This array is capable of simultaneously or sequentially receiving light signals of multiple wavelengths reflected from wound tissue and converting them into electrical signals.

[0041] Organic photodetectors are photoelectric conversion devices made using organic semiconductor materials. Compared to inorganic photodetectors, organic photodetectors are characterized by their flexibility, ability to be fabricated over large areas, tunable wavelength response, and high integration with flexible substrates.

[0042] Light source 202 is used to emit light of a specific wavelength to irradiate wound tissue. In this embodiment, light source 202 can be an organic light-emitting diode (OLED) light source, which is characterized by customizable wavelength, fast response speed, and easy integration onto a flexible substrate.

[0043] The positioning area 104 is a region on the upper surface of the flexible substrate 101 used to guide the user to place the wound area at the detection position. The setting of this area helps to ensure the consistency of the relative position between the wound and the photoelectric detection array 201 during each detection.

[0044] Pressure sensor 301 is used to detect the contact pressure between the device and a finger or limb. By monitoring the pressure, posture parameters can be obtained for subsequent correction of optical reflectivity data.

[0045] An angle sensor 302 is used to detect the spatial posture of the device or finger pressing direction. This sensor can provide pressing angle information, which, together with pressure data, constitutes posture parameters to compensate for detection errors caused by posture changes.

[0046] The processing unit 401 is the control and data processing module of the device, and typically includes a microprocessor. This microprocessor integrates an analog-to-digital converter, a storage unit, and an embedded computing unit. It is responsible for controlling the emission of light from the light source 202, collecting signals from the photoelectric detection array 201, the pressure sensor 301, and the angle sensor 302, and performing data processing, analysis, and result output.

[0047] A flexible substrate 101 is designed as the support structure for the device, on which a photodetector area 102 and a light source area 103 are integrated. The flexible substrate 101 can be made of a flexible polymer material, such as polyethylene terephthalate (PET) film or polyethylene naphthalate (PEN) film. This flexibility allows the device to adapt to irregularities on the human skin surface, such as finger joints or limb bends, thereby maintaining contact during detection. The photodetector area 102 is typically located at the center of the flexible substrate 101 and is used to receive light signals reflected from wound tissue. The light source area 103 is arranged around the photodetector area 102 and is used to mount multiple light sources 202 to provide illumination of the wound tissue.

[0048] A photodetector array 201 is disposed within the photodetector region 102, and its core component is at least one organic photodetector (OPD). As one implementation, the photodetector array 201 can consist of a single large-area organic photodetector capable of covering the entire photodetector region 102 to receive reflected light over a large range. Alternatively, the photodetector array 201 can be composed of multiple discrete organic photodetector units arranged in a matrix, each unit operating independently to provide spatial resolution. Organic photodetectors are chosen for their inherent flexibility, customizable wavelength response, and compatibility with the flexible substrate 101, enabling efficient conversion of reflected light signals into electrical signals.

[0049] Multiple light sources 202 are arranged within the light source area 103, spaced apart along a circumferential direction. The emitting surface of each light source 202 is designed to face the photoelectric detection area 102 to ensure that light can effectively illuminate the wound tissue located above the device. These light sources 202 are used to emit light sequentially at multiple preset wavelengths. For example, different colored light-emitting diodes (LEDs) can be used as light sources 202, and multi-wavelength illumination can be achieved by controlling their sequential emission. The purpose of this multi-wavelength illumination is to obtain the reflectance characteristics of the wound tissue under different spectra, thereby providing optical information. The circumferential arrangement of the light sources 202 helps to achieve uniform illumination of the wound tissue and reduce detection errors caused by uneven illumination.

[0050] A positioning area 104 is disposed on the upper surface of the flexible substrate 101. The planar projection of the positioning area 104 on the flexible substrate 101 at least partially overlaps with the photoelectric detection area 102. The function of the positioning area 104 is to guide the user to align the wound area of ​​the finger or limb with the photoelectric detection array 201. As one implementation, the positioning area 104 can be a graphic mark printed on the surface of the flexible substrate 101, such as a circular or square outline, which the user visually aligns with the wound. Alternatively, the positioning area 104 can be an area with a specific texture or color to help the user locate the wound through visual cues.

[0051] At least one pressure sensor 301 is disposed within the positioning area 104 for detecting the contact pressure between a finger or limb and the device. As one implementation, the pressure sensor 301 can be a thin-film pressure sensor capable of sensing changes in pressure and outputting a corresponding electrical signal. Alternatively, the pressure sensor 301 can be a piezoresistive sensor, whose resistance changes with the applied pressure. By monitoring the contact pressure, posture parameters can be obtained, providing a basis for subsequent correction of optical reflectivity data.

[0052] An angle sensor 302 is disposed on the flexible substrate 101 and / or the circuit board below it to detect the spatial attitude of the device or finger pressing direction. As one implementation, the angle sensor 302 can be a single-axis or multi-axis accelerometer that infers the tilt angle of the device by measuring changes in the direction of gravity. Alternatively, the angle sensor 302 can be a gyroscope that detects the angular velocity of the device and then calculates the attitude angle. The angle sensor 302 works in conjunction with the pressure sensor 301 to provide attitude information to compensate for the impact of changes in the pressing angle on reflectivity detection.

[0053] The processing unit 401 includes a microprocessor that integrates an analog-to-digital converter, a storage unit, and an embedded computing unit. The processing unit 401 is electrically connected to the photodetector array 201, the pressure sensor 301, the angle sensor 302, and the light source 202. The processing unit 401 coordinates the operation of each component, such as controlling the light source 202 to emit light in a preset sequence and wavelength, while simultaneously acquiring reflected photoelectric signals from the photodetector array 201. Furthermore, the processing unit 401 is also responsible for acquiring contact pressure and spatial attitude data from the pressure sensor 301 and the angle sensor 302. The analog-to-digital converter converts analog signals into digital signals, the storage unit stores the acquired raw data and results, and the embedded computing unit performs data processing, attitude correction algorithms, and computational tasks.

[0054] The following example will provide a more detailed explanation of the above technical solution: Suppose User A needs to regularly monitor the healing of a wound on their finger at home. User A takes out the device and prepares to perform the test. First, User A places the injured finger on the flexible substrate 101 of the device. The flexibility of the substrate 101 allows the device to conform to the curvature of the finger, ensuring contact between the detection area and the device. User A aligns the wound area with the photodetector array 201 above the photodetector detection area 102 by observing the indication in the positioning area 104.

[0055] During contact between the finger and the device, a pressure sensor 301 located in the positioning area 104 detects the contact pressure applied by the finger to the device. Simultaneously, an angle sensor 302 located on the flexible substrate 101 detects the spatial posture of the finger pressing the device, such as the tilt angle. This pressure and angle information is transmitted to the processing unit 401. The embedded computing unit in the processing unit 401 uses this posture data, combined with a stored "pressure-angle-reflectivity" correction model, to calibrate subsequently acquired optical reflectivity data, thereby eliminating errors caused by changes in pressure or angle.

[0056] Once the posture detection meets the preset conditions, the processing unit 401 controls multiple light sources 202 within the light source area 103 to emit light sequentially according to a preset order and wavelength, illuminating the wound tissue on the finger wound surface. For example, the light sources 202 can sequentially emit light with wavelengths of 450nm, 550nm, 650nm, and 950nm. After these lights penetrate the wound tissue, they are partially reflected and received by the photodetector array 201 within the photodetector area 102. The organic photodetector in the photodetector array 201 converts the received multi-wavelength reflected light into electrical signals, which are then converted into digital signals by an analog-to-digital converter and subsequently transmitted to the processing unit 401.

[0057] After receiving the multi-wavelength reflected light, the processing unit 401 first calls the correction model stored in the storage unit. Using the previously collected pressure and angle data, it corrects the original reflectivity to obtain a corrected reflectivity equivalent to that under standard posture. These corrected multi-wavelength reflectivities are used as feature inputs and fed into the wound quantification correlation model running in the processing unit 401. This model is trained based on clinically labeled data and can map optical reflectivity features to healing levels. For example, the model may output healing levels such as "granulation tissue growth phase" or "epidermal repair phase".

[0058] Furthermore, the processing unit 401 will combine user A's historical detection data and / or individual physiological characteristics (such as age, blood sugar level, etc.) to invoke the stored healing time prediction model. This model will comprehensively analyze the current healing level, multi-wavelength reflectance change trends, and individual characteristics to estimate the remaining time for wound healing and provide warnings of delayed healing or infection risk. Finally, these healing level, estimated healing time, and risk warning results will be output through the device's display module or a connected external terminal, allowing user A to understand the healing status of their wound. Simultaneously, all data from this detection, including multi-wavelength reflectance, posture parameters, and prediction results, will be stored in the local storage unit or uploaded to the cloud for subsequent continuous monitoring and model optimization.

[0059] Traditional optical reflection detection devices, such as those mentioned in the background art, often suffer from inconsistent contact postures between the user's finger or limb and the probe during detection. Changes in pressure and angle alter the paths of incident and reflected light, leading to fluctuations in reflectivity readings. The lack of pressure / angle calibration and compensation mechanisms further impacts the accuracy of the detection results. However, in this embodiment, by placing a pressure sensor 301 within the positioning area 104 and an angle sensor 302 on the flexible substrate 101 and / or its underlying circuit board, and electrically connecting them to the processing unit 401, the device can acquire the user's contact pressure and spatial posture information during pressing. The processing unit 401 uses these posture parameters to correct the original reflectivity, effectively eliminating errors introduced by posture changes. For example, in the detection of user A described above, even with deviations in the force or angle of finger pressing, the processing unit 401 can adjust the original reflectivity to a value equivalent to the standard posture using a correction model, thus ensuring the stability and accuracy of the detection results—a capability lacking in existing technologies.

[0060] Furthermore, existing optical reflectance detection devices generally fail to integrate with artificial intelligence algorithms such as deep learning, making it difficult to comprehensively analyze multi-wavelength reflectance characteristics and individual physiological parameters to provide quantitative results for wound healing levels, let alone predict the remaining healing time. This embodiment, however, utilizes an embedded computing unit integrated into the processing unit 401 to achieve the ability to analyze multi-wavelength reflectance data. The processing unit 401 can not only take corrected multi-wavelength reflectance as input and output the healing level through a pre-trained wound quantification correlation model, but also combine historical data and individual physiological characteristics to invoke a healing time prediction model, providing the remaining healing time and risk warnings. This mechanism of integrating optical detection with intelligent algorithms elevates wound healing assessment from raw data reading to quantification, prediction, and risk assessment, improving the intelligence and practicality of the assessment and providing users with health management information.

[0061] Furthermore, this embodiment also optimizes the device structure. In the prior art, the combination of OLED light source and OPD array in the field of wound detection is limited, and there is a lack of flexible modules with an integrated layout of ring-shaped multi-wavelength emission and central array detection for small-area wounds such as fingers. This embodiment uses a flexible substrate 101, on which a photoelectric detection area 102 and a light source area 103 are formed. A photoelectric detection array 201 (including organic photodetectors) is set in the photoelectric detection area 102, and multiple light sources 202 (such as OLED light sources) are arranged at intervals along the circumference in the light source area 103. This flexible, integrated, ring-shaped multi-wavelength emission and central array detection layout allows the device to fit closely to small-area wounds, achieving close-range, multi-wavelength, and highly integrated optical reflection detection, reducing the size and thickness of the device, making it more suitable for home or wearable applications, and overcoming the limitations of traditional medical imaging equipment that are large, costly, and inconvenient for home use.

[0062] In summary, this embodiment effectively solves the problems of optical reflection detection being susceptible to attitude interference and lacking intelligent prediction by introducing an attitude sensor and an intelligent processing unit. At the same time, through flexible and integrated device design, it improves the portability and applicability of the device, providing an innovative technical solution for wound healing status detection and healing time prediction in home scenarios.

[0063] Example 2, A device for detecting wound healing status and predicting healing time based on optical reflectivity, such as... Figure 1 and Figure 2 As shown, it includes: A flexible substrate on which a photoelectric detection area and a light source area surrounding the photoelectric detection area are formed; A photoelectric detection array is set in the photoelectric detection area, and the photoelectric detection array includes at least one organic photodetector; Multiple light sources are arranged at intervals along the circumference within the light source area, with the emitting surface of each light source facing the photoelectric detection area, and are used to sequentially irradiate the wound tissue located above the device with multiple preset wavelengths. A positioning area is set on the upper surface of a flexible substrate, and the planar projection of the positioning area on the flexible substrate at least partially overlaps with the photoelectric detection area, for guiding the wound area of ​​the finger or limb to align with the photoelectric detection array. At least one pressure sensor is disposed in the positioning area for detecting the contact pressure between the finger and the device; An angle sensor, mounted on a flexible substrate and / or a circuit board beneath it, is used to detect the spatial orientation of the device or finger pressing direction. The processing unit includes a microprocessor integrating an analog-to-digital converter, a storage unit, and an embedded computing unit. The processing unit is electrically connected to a photodetector array, a pressure sensor, an angle sensor, and a light source. The flexible substrate is a polyimide substrate covered with a medical-grade elastic encapsulation layer. The encapsulation layer forms a positioning area on its upper surface for direct contact with the skin and to provide flexible adhesion.

[0064] The flexible substrate 101 serves as the core physical support for the entire detection device, supporting key components such as the photoelectric detection area 102, the light source area 103, the photoelectric detection array 201, multiple light sources 202, the positioning area 104, the pressure sensor 301, the angle sensor 302, and the processing unit 401. To address the issues of insufficient flexibility and biocompatibility in traditional substrates when in contact with skin, leading to poor adhesion, low comfort, and consequently affecting detection accuracy and user experience, this application optimizes the flexible substrate 101. Specifically, the flexible substrate 101 is designed as a polyimide substrate, possessing excellent mechanical strength and flexibility. This provides stable support for precision electronic components such as the photoelectric detection array 201 and the light source 202 within the device, while allowing the substrate to bend and deform as a whole to adapt to the complex curvature of wound areas on human fingers or limbs. Above this polyimide substrate, a medical-grade elastic encapsulation layer 405 is applied. This encapsulation layer 405 not only provides additional physical and chemical protection for the internal components but, more importantly, serves as the interface between the device and the skin. The introduction of a medical-grade elastic encapsulation layer 405 significantly improves the device's biocompatibility, reduces skin irritation, and provides a soft touch, thereby greatly enhancing user comfort during wear or use. The elastic encapsulation layer 405 further forms a positioning area 104 on its upper surface. The positioning area 104 is designed to provide clear physical or visual guidance to the user, ensuring that the wound area of ​​the finger or limb is accurately aligned with the photodetector array 201 inside the device. In this way, the light emitted by the light source 202 can effectively illuminate the wound tissue to the greatest extent possible, and the reflected light can be accurately received by the photodetector array 201, thereby reducing measurement errors caused by inaccurate contact posture. The combination of the medical-grade elastic encapsulation layer 405 and the positioning area 104 allows the device to achieve a flexible fit with the skin. This flexible fit not only reduces the gap between the device and the skin, lowering the possibility of ambient light interference, but also ensures that the device maintains a stable contact state even with slight movement of the finger or limb during the detection process, thereby improving the accuracy and reliability of optical reflectivity measurements. Therefore, by using polyimide as a flexible substrate 101, and covering it with a medical-grade elastic encapsulation layer 405, with a positioning area 104 formed on its upper surface, the entire device maintains its core optical detection function while greatly enhancing its adhesion to human skin, comfort, and biocompatibility. This structural optimization ensures that the device can provide more stable and accurate detection data in practical applications, laying a solid foundation for the subsequent processing unit 401 to assess wound healing status and predict healing time.

[0065] The following is a specific example. The flexible substrate 101 of this application can be a polyimide film with a thickness of approximately 75 micrometers, such as the Kapton HN series products from DuPont. This thickness of polyimide film provides sufficient mechanical strength while maintaining good flexibility, allowing it to be easily bent to adapt to the curvature of fingers or limbs. On top of the polyimide substrate 101, a medical-grade elastic encapsulation layer 405 with a thickness of approximately 1 millimeter can be applied. This encapsulation layer 405 can be formed from transparent medical-grade silicone rubber (such as Shin-Etsu Chemical's KE-1950) through a molding process. Silicone rubber has excellent biocompatibility, is unlikely to cause skin allergic reactions, and has a moderate elastic modulus, providing a soft touch and good cushioning effect. When the positioning area 104 is formed on the upper surface of the encapsulation layer 405, a groove can be pre-set in the mold, so that after the silicone rubber cures, a circular recessed area with a diameter of approximately 15 millimeters is formed at the corresponding position. This circular recessed area serves as a physical guide, allowing users to place their injured fingertip or wound within it. Furthermore, a textured surface with micron-level bumps can be formed at the edge of this recessed area using laser etching to further enhance the tactile guidance effect, helping users more accurately align the wound area with the photoelectric detection array 201 below. This design enables the device to achieve a close and comfortable flexible fit against the skin, thereby ensuring the stability and accuracy of optical detection.

[0066] In this embodiment, the light source has a preset tilt angle relative to the flexible substrate 101 to form annular multi-wavelength oblique incidence illumination. The "preset tilt angle" refers to a non-zero fixed angle between the light source 202's luminous axis and the surface normal of the flexible substrate 101 during installation or design. This tilt angle design ensures that the light does not incident perpendicularly to the wound surface, but rather at a certain angle. This tilt angle can be achieved by designing a mounting structure with an inclined surface on the flexible substrate 101, allowing the light source 202 to be embedded or fitted at a specific angle; or by integrating micro-optical elements, such as prisms or tilted microlens arrays, above the luminous surface of the light source 202 to change the light emission direction, thereby achieving the preset oblique incidence effect overall. "Annular multi-wavelength oblique incidence illumination" refers to multiple light sources 202 arranged along a circular path or annular region, surrounding the photodetector area 102. This arrangement ensures uniform and multi-directional illumination of the wound tissue in the central area. "Multi-wavelength" means that these light sources 202 can emit light of different wavelengths, for example, by using different types of OLED light sources 202 or by emitting light of different wavelengths at different times using a single OLED light source 202. "Oblique incidence illumination" combines the aforementioned preset tilt angle so that the multiple light sources 202 arranged in a ring illuminate the wound tissue in an oblique manner.

[0067] In one specific implementation, the light source 202 can be a micro-OLED light source array, which is integrated within the light source region 103 of the flexible substrate 101. To achieve a preset tilt angle, the flexible substrate 101 can pre-form a series of micro-sloping structures at corresponding positions in the light source region 103 using molding or etching processes. The OLED light sources 202 are then directly attached to or encapsulated on these micro-sloping surfaces, thereby naturally forming an angle of inclination towards the photodetector region 102. For example, the light-emitting surface of each OLED light source 202 can form an angle of approximately 15 to 30 degrees with the plane of the flexible substrate 101. These OLED light sources 202 are uniformly arranged in a ring and can be designed to emit light of different wavelengths at different times. For example, the first group of OLED light sources 202 emits light with a wavelength of 450 nm, the second group emits light with a wavelength of 550 nm, and so on, covering multiple preset wavelength ranges. When the device is in operation, these ring-shaped OLED light sources 202 with preset tilt angles will irradiate the wound tissue in a way that is obliquely incident, either sequentially or simultaneously, and the reflected light will be received by the photodetector array 201 composed of organic photodetectors located at the center.

[0068] The flexible substrate 101, corresponding to the photoelectric detection area 102, has a groove at its center. This groove is a specific shaped area formed by downward indentation on the surface of the flexible substrate 101, and its main function is to accommodate and fix the photoelectric detection array 201, precisely positioning it in the central region of the flexible substrate 101. This groove can be achieved in various ways, such as by directly forming it on the flexible substrate 101 through molding, processing it on the flexible substrate 101 through laser etching, chemical etching, or by attaching or laminating an additional flexible layer with a pre-formed groove onto the flexible substrate 101. Simultaneously, the flexible substrate 101, corresponding to the light source area 103, has an annular groove surrounding the outside of the photoelectric detection area 102. This annular groove is a recessed area distributed in a ring around the central groove (i.e., the photoelectric detection area 102), and its function is to accommodate and fix multiple light sources 202, arranging them evenly along the circumference to form an annular lighting structure. The annular groove can be formed simultaneously with the central groove through an integrated molding process, or it can be formed in stages, such as first forming the central groove and then forming the annular groove on its outer side through precision machining, or by stacking multiple layers of materials and selectively removing or stacking them.

[0069] The light sources are uniformly arranged circumferentially along the annular light source region 103, and include at least four groups of OLED light sources with emission wavelengths covering approximately 450–950 nm and single-device light intensity of approximately 0.1–10 mW / cm². 2The response time is less than 10 μs. The uniform circumferential arrangement of the light sources along the annular light source region 103 means that the light sources 202 are arranged at equal intervals or symmetrically within the annular region formed around the photodetector region 102 on the flexible substrate 101. This arrangement aims to ensure that light is uniformly irradiated onto the wound tissue located above the photodetector array 201 from all directions, avoiding reflectivity measurement deviations caused by uneven local irradiation. For example, the light sources 202 can be arranged at equal angular intervals in a circular array on the light source region 103, or arranged at the vertices of symmetrical quadrilaterals or hexagons to achieve circumferential uniformity. The light sources include at least four sets of OLED light sources, where OLED light sources are organic light-emitting diode light sources with characteristics such as self-illumination, flexibility, thinness, and customizable wavelength. The setting of at least four sets of OLED light sources aims to provide sufficient light source quantity and redundancy, ensuring sufficient coverage and light intensity during multi-wavelength irradiation, and improving the robustness of the system. Each OLED light source group can consist of one or more OLED light-emitting units, which can be integrated on the same chip or used as independent devices, or grouped according to different wavelength requirements. The emission wavelength coverage of approximately 450–950 nm refers to the range of light wavelengths emitted by the OLED light source 202, extending from the blue region of the visible light spectrum to the near-infrared region. This broad wavelength range enables the detection of the optical absorption and scattering characteristics of biological tissues at different wavelengths, thereby obtaining richer tissue information, such as changes in the content of components like hemoglobin, water, and melanin, which are closely related to wound healing status. This wavelength coverage can be achieved by using OLED devices with different material systems or dopants, or by integrating multiple narrowband OLED light sources together. The light intensity of a single device is approximately 0.1–10 mW / cm². 2 The light intensity refers to the light intensity emitted by a single OLED light source 202 device during operation. This light intensity is designed to provide sufficient light energy to penetrate wound tissue and generate a reflected signal that can be effectively received by the photodetector array 201, while avoiding excessive light intensity that could cause thermal or photodamage to the tissue. This can be achieved by optimizing the structural design, material selection, and driving current of the OLED device, or by precisely controlling it by adjusting the driving voltage or current. The response time of less than 10 μs refers to the time required for the OLED light source 202 to emit a stable light signal from receiving the driving signal, or the time required for the light signal to attenuate to a preset threshold from the off signal. A fast response time ensures that the light source 202 can quickly and accurately switch between different wavelengths, adapt to multi-wavelength sequential illumination detection modes, reduce crosstalk between different wavelength signals, and improve the efficiency and accuracy of data acquisition. This can be achieved by using high-mobility organic semiconductor materials, optimizing the device structure, and designing efficient driving circuits.

[0070] In one specific implementation, the flexible substrate 101 can be made of polyimide (PI) material, and the light source region 103 formed thereon can be an annular region with a diameter of approximately 20 mm. Within this annular light source region 103, six groups of OLED light sources 202 can be arranged, each group containing one or two OLED light-emitting units. These OLED light sources 202 can be uniformly arranged at 60-degree intervals along the circumference of the annular light source region 103, for example, at positions of 0°, 60°, 120°, 180°, 240°, and 300° respectively. To achieve wavelength coverage of 450–950 nm, six OLED light sources 202 can be configured to emit light with wavelengths of 450 nm (blue), 520 nm (green), 660 nm (red), 780 nm (near-infrared), 850 nm (near-infrared), and 940 nm (near-infrared), respectively. The driving circuit of each OLED light source 202 can be designed to provide a light intensity of approximately 5 mW / cm² to ensure sufficient penetration depth and signal strength. The material system of the OLED light source 202 can be selected from phosphorescent or fluorescent materials, combined with a microcavity structure to achieve a fast response of less than 10 μs. The processing unit 401 controls the driving circuit of the OLED light source 202 to realize the rapid sequential switching on and off of light sources of different wavelengths. For example, the OLED light sources 202 of different wavelengths are switched on and off sequentially at 100 μs intervals, while the photodetector array 201 synchronously collects the reflected signals.

[0071] As a specific example of the above embodiments, the device employs a flexible polyimide (PI) substrate 101 with a thickness of approximately 0.2 mm, such as... Figure 1As shown, the following semiconductor processes are sequentially formed on the substrate through photolithography, etching, and vapor deposition: A photodetector area 102, formed by a central circular groove, is used to place an OPD array, with a diameter of approximately 6 mm. An annular groove, i.e., the light source area 103, is located outside the central groove and is used to embed an OLED light source, with an inner diameter of approximately 6 mm and an outer diameter of approximately 10 mm. A positioning area 104, formed by an outer arc-shaped positioning groove, guides the user's finger placement, with an arc radius of approximately 10 mm and a groove depth of approximately 2 mm. The user naturally places the injured finger in the arc-shaped positioning groove, with the finger's pad covering the corresponding positions of the central OPD area and the annular OLED area, thereby defining the measurement area and improving the consistency of the measurement posture. Four OLED light sources 201 are embedded within the annular groove 103, arranged at approximately equal intervals along the circumference, each with an angle of approximately 15° relative to the substrate normal, and their emitting surfaces facing the central area. The OLED is preferably a blue-enhanced organic light-emitting device (OLED), with an emission wavelength covering approximately 450–950 nm, a single-device light intensity of approximately 0.1–10 mW / cm², and a response time of less than 10 μs. The emissive layer can be an organic light-emitting structure doped with Alq3 and Ir(ppy)3. A 3×3 OPD detector array 202 is disposed within the central recess, with the photosensitive surface of the OPD essentially flush with the substrate surface. The response wavelength range of a single OPD is approximately 400–1000 nm, the detection sensitivity is not less than 0.1 μA / μW, and the dark current is less than approximately 1 nA. A copper phthalocyanine / fullerene (CuPc / C60) organic heterojunction structure is preferred. Figure 1 As shown, the attitude detection module is located inside the arc-shaped positioning groove and includes: two miniature piezoresistive pressure sensors 301, preferably attached to the bottom or side wall of the positioning groove, with a measurement accuracy of approximately 0.1 N, used to detect the contact pressure when the user's finger presses the device, with a response time of less than 50 ms; and a MEMS angle sensor 302, preferably deployed under a flexible substrate or on a supporting structure, with an angle measurement accuracy of approximately 1°, used to detect the attitude angle of the device relative to the horizontal plane or the angle change of the finger pressing direction. Through the pressure sensor 301 and the angle sensor 302, the pressing force and pressing angle at each detection can be acquired in real time and used as input for the subsequent reflectivity correction model. Figure 1As shown, the processing and storage module includes: a microprocessor 401, preferably an ARM Cortex-M4 core microcontroller, integrating at least a 12-bit analog-to-digital converter (ADC) for acquiring OPD output electrical signals and attitude sensor signals, and performing operations such as filtering, normalization, reflectivity calculation, and correction model calling; and a memory for storing parameters of the "pressure-angle-reflectivity" correction model, parameters of the wound quantification level correlation model, AI prediction model weights, and historical detection data. The microprocessor 401 is electrically connected to the optical emission and detection module and the attitude detection module via a flexible printed circuit board (FPC). This system is used to perform the following functions: acquire OPD output electrical signals and attitude sensor signals; filter and normalize the acquired signals, and calculate multi-wavelength reflectance; call the pre-stored "pressure-angle-reflectance" correction model to compensate for the original reflectance and obtain corrected reflectance data; store and update the parameters of the "pressure-angle-reflectance" correction model, the parameters of the wound quantification level association model, and the weight parameters of the healing time prediction model; store historical data for each detection, including multi-wavelength corrected reflectance, attitude parameters, healing level, and prediction results; deploy and run the lightweight fusion model, output the wound healing level based on the multi-wavelength corrected reflectance, and output the remaining healing time and risk warnings by combining the reflectance time series and individual physiological characteristics.

[0072] The entire outer surface of the module is encapsulated with approximately 0.5 mm thick medical-grade PDMS (polydimethylsiloxane) 405, forming a waterproof, dustproof, and skin-contact-safe encapsulation layer, achieving a protection level similar to IP67. The device can be further integrated with: a small display unit 501 (such as an OLED screen) to display the current healing level, estimated healing time, and prompts; and indicator lights and / or a buzzer 502 to provide audible and visual prompts such as incorrect posture or completion of the test.

[0073] Example 3, A prediction method based on the above-mentioned wound healing status detection and healing time prediction device, the device being a portable detection device equipped with an AI processing module, includes the following steps: Step A, Posture Guidance and Contact Detection: The user places the injured finger or other wound on the positioning area of ​​the device. The pressure sensor detects whether the contact pressure is within the preset range, and the angle sensor detects whether the pressing angle is within the preset range. When the pressure and angle meet the set conditions, a contact status compliance prompt is output, allowing the process to proceed to the optical detection stage. Otherwise, the user is prompted to adjust the pressing posture through an audio-visual signal. Step B: Multi-wavelength optical illumination and reflection signal acquisition; control the OLED light source of each wavelength to emit light in a preset order, and acquire the multi-wavelength reflected photoelectric signals output by the photoelectric detection array as well as the corresponding pressure and angle information; according to the signal processing and the pre-stored "pressure-angle-reflectivity" correction model, correct the original reflectivity obtained under different pressures and angles to the corrected reflectivity equivalent to the standard posture. Step C: Output of wound healing quantification level; The corrected multi-wavelength reflectance is used as input and fed into the wound quantification correlation model trained with clinical annotation data to output the corresponding wound healing level; The healing level is divided into multiple levels according to preset standards to characterize different healing stages from superficial repair, granulation growth, deep healing to severe damage. Step D: Healing time prediction and risk assessment; Based on historical test data and / or individual physiological characteristics, call the pre-stored healing time prediction model to output the remaining wound healing time and / or healing delay risk warning; Step E, Result Display and Data Storage: The current wound healing level, estimated healing time, and risk warnings are visualized and output through the display module or external terminal. At the same time, information such as the multi-wavelength reflectivity, attitude parameters, and prediction results of this test are stored locally or in the cloud for subsequent continuous monitoring and incremental model optimization.

[0074] In the above method, the posture guidance and contact detection in step A are designed to ensure the standardization of the detection process and the accuracy of data acquisition. The user places the injured finger or other wound on the positioning area 104 of the device. This positioning area 104 typically has a specific shape or marking to guide the user to accurately align the wound area with the photoelectric detection array 201. The pressure sensor 301 is used to monitor the force of the user pressing the device in real time, and its detection result is compared with a preset pressure range to determine whether the pressure is appropriate. For example, the pressure sensor 301 can be a resistive thin-film pressure sensor or a capacitive pressure sensor, which quantifies the pressure by measuring the change in resistance or capacitance caused by deformation. The angle sensor 302 is used to detect the user's spatial posture when pressing the device, ensuring that the pressing angle meets the preset requirements and avoiding optical path deviation due to excessive tilt. The angle sensor 302 can be a microelectromechanical system (MEMS) inertial measurement unit (IMU) or a tilt sensor, which determines the pressing angle by measuring the angle between the direction of gravity and the plane of the device. When both pressure and angle meet the preset conditions, the system will output a prompt indicating that the contact status has met the requirements, such as displaying "Good posture" on the screen or emitting a short beep, thus allowing the process to proceed to the subsequent optical inspection. Conversely, if the conditions are not met, the system will prompt the user to adjust the pressing posture through audio-visual signals, such as flashing LEDs or a continuous beep, until the standard is met.

[0075] Step B, multi-wavelength optical irradiation and reflection signal acquisition, is a crucial step in obtaining wound optical information. In this step, the processing unit 401 controls the OLED light source 202 of various wavelengths to emit light sequentially in a preset order. For example, it can emit light at a wavelength of 450nm first, followed by 550nm, 650nm, etc., ensuring sufficient irradiation of the wound tissue at different wavelengths. The OLED light source 202 has the advantage of customizable wavelengths, covering multiple bands from visible light to near-infrared to obtain richer tissue optical information. Simultaneously, the photoelectric detection array 201 synchronously acquires multi-wavelength photoelectric signals reflected back from the wound tissue and converts them into electrical signals. The photoelectric detection array 201 can be composed of multiple organic photodetectors, which are responsive to different wavelengths of light. During this process, the system also synchronously records real-time pressure and angle information output by the pressure sensor 301 and angle sensor 302. These raw reflectivity data and corresponding pressure and angle information are then sent to the pre-stored "pressure-angle-reflectivity" correction model in the processing unit 401 for processing. The purpose of this correction model is to eliminate the influence of changes in user pressure and angle on reflectivity measurement results, and to correct the original reflectivity obtained under different postures to the corrected reflectivity equivalent to that under the standard posture, thereby ensuring the accuracy and comparability of the data.

[0076] The wound healing quantification level output in step C aims to convert optical signals into clinically understandable healing status. The multi-wavelength reflectance data corrected in step B is used as input and fed into a pre-stored wound quantification correlation model in processing unit 401. This model, trained on a large amount of clinically labeled data, establishes a non-linear mapping relationship between multi-wavelength reflectance features and wound healing level. The model outputs multiple levels according to preset standards, such as levels 1-10, to characterize different healing stages from superficial repair, granulation tissue growth, deep healing to severe injury. This quantified level output makes the assessment of wound healing status more objective and precise.

[0077] Step D, healing time estimation and risk assessment, provides a prediction of the future healing trend of the wound. In making the estimation, the system comprehensively considers historical test data (if available) and / or individual physiological characteristics (such as age, blood sugar levels, underlying diseases, etc.). Processing unit 401 calls a pre-stored healing time prediction model, which uses advanced machine learning algorithms to analyze the current wound healing level, multi-wavelength reflectance variation trends, and individual physiological parameters, thereby outputting the remaining wound healing time (e.g., how many more days are expected to take for complete healing) and / or a healing delay risk warning (e.g., a warning of infection risk or slower-than-expected healing speed).

[0078] Step E, result display and data storage, is the final stage of the entire detection process. The wound healing level, estimated healing time, and risk warnings obtained from the current detection will be visualized through a display module (such as the device's built-in OLED screen) or an external terminal (such as a smartphone app connected via Bluetooth), allowing users or medical personnel to intuitively understand the wound condition. Simultaneously, all key data from this detection, including multi-wavelength reflectance, posture parameters (pressure, angle), and prediction results, will be stored in local storage or uploaded to a cloud server. This stored data is crucial for subsequent continuous monitoring, healing trend analysis, and incremental model optimization, providing data support for long-term management and personalized treatment.

[0079] This application's solution integrates a series of steps, including attitude guidance and contact detection, multi-wavelength optical illumination and reflection signal acquisition, wound healing quantification level output, healing time prediction and risk assessment, and result display and data storage, to form a complete method for wound healing status detection and healing time prediction. This method fully utilizes hardware components integrated on the flexible substrate 101, such as the photoelectric detection array 201, OLED light source 202, pressure sensor 301, and angle sensor 302, and achieves full automation from raw signal acquisition to intelligent prediction through the correction model and AI prediction model embedded in the processing unit 401.

[0080] Example 4, A prediction method based on the wound healing status detection and healing time prediction device of Embodiment 2 above, such as... Figures 1 to 6 As shown, the portable detection device equipped with an AI processing module includes the following steps: Step A: Posture Guidance and Contact Detection; The user places the injured finger into the arc-shaped positioning groove 104, and the device initiates posture detection. The pressure sensor detects whether the current contact pressure is within a preset range (e.g., approximately 0.8–1.2 N), and the angle sensor detects whether the current pressing angle is within a preset range (e.g., ±5°). When both pressure and angle meet the conditions, the indicator light illuminates green or displays "Normal Posture," and the process proceeds to the next step; otherwise, the user is prompted by sound and light to adjust their finger posture. Step B: Multi-wavelength optical illumination and reflection signal acquisition; The microprocessor controls four OLED light sources to emit light sequentially according to a predetermined order and wavelength, for example, emitting light at wavelengths of approximately 450 nm, 660 nm, and 850 nm, illuminating the wound area. The central OPD array synchronously acquires the reflected light signals at each wavelength and converts them into electrical signals. After sampling by an ADC, the raw reflectivity data R is obtained. 原始The microprocessor reads the currently detected pressure P and angle θ, calls the "pressure-angle-reflectivity" correction model to calculate the correction factor, compensates for the original reflectivity at each wavelength, and obtains the corrected reflectivity R.

[0081] In step B, the specific steps for constructing the "pressure-angle-reflectivity" correction model are as follows: 1) Select standard samples with smooth surfaces and stable reflectivity, such as polished quartz sheets, as calibration objects and fix them in the detection area of ​​the device.

[0082] 2) Within the preset pressure and angle range, design several pressure-angle combination points. At each combination point, apply the corresponding pressure and angle through the device, measure the original reflectivity Roriginal of multiple wavelengths multiple times, take the average value, and record the corresponding pressure P and angle θ to form a calibration dataset. 3) The reference reflectance R is measured under standard pressure and standard angle conditions. A simplified correction formula is obtained by fitting the deviations of each combination point through polynomial regression, for example: Rcorrected = Roriginal × (0.98 + 0.01P - 0.002θ) Where P is the pressure applied and θ is the pressure angle.

[0083] 4) The error between the corrected reflectance and the reference reflectance is evaluated by cross-validation or partitioning the validation set. When the error meets the preset index, the corrected model parameters are stored in the microprocessor. In actual testing, the microprocessor calculates P and θ based on the real-time output of the pressure sensor and angle sensor, and calls the above correction formula to compensate for the original reflectivity, thus obtaining the target reflectivity R correction after attitude correction.

[0084] The "standard sample with a smooth surface and stable reflectivity, such as a polished quartz plate" refers to a material sample with known and stable reflectivity characteristics at different wavelengths. Its function is to provide a reliable benchmark for evaluating and calibrating the measurement deviation of the device under different contact postures. Besides polished quartz plates, other options include precision-machined ceramic standard plates, glass plates with specific optical coatings, or certified diffuse reflection standard white plates. These samples should ensure high surface flatness to avoid measurement errors introduced by uneven sample morphology, and their reflectivity should have good stability within the device's operating wavelength range. "Fixing it in the detection area of ​​the device" means placing or mounting the aforementioned standard sample in a stable and repeatable manner above the photoelectric detection area 102 of the device, allowing it to be illuminated by the light source 202 and received by the photoelectric detection array 201. Fixing methods may include, but are not limited to: mechanical fixation using precision clamps, adsorbing the sample at a specific position using a vacuum adsorption device, or precisely inserting the sample using a repeatable positioning slot structure. This fixation aims to ensure the accuracy and consistency of the sample position during calibration, thereby eliminating errors introduced by sample position variations.

[0085] "Preset pressure and angle ranges" refers to the need to predetermine a range of pressure and angle values ​​that covers actual usage scenarios when building the correction model. For example, the pressure range can be set based on the typical force of a user's finger or limb pressing, and the angle range can be set based on the possible tilt angle of the user's press. These ranges can be determined based on ergonomic research, clinical experience, or prior user test data to ensure the model effectively covers various contact postures that may be encountered in actual testing. "Designing several pressure-angle combination points" refers to strategically selecting a series of discrete pressure and angle value pairs within the aforementioned preset pressure and angle ranges to form test points for calibration. These combination points can use a uniformly distributed grid sampling method, such as dividing the pressure and angle ranges into several equal segments and then taking their intersections; or a more complex sampling strategy, such as Latin hypercube sampling, can be used to more effectively cover the parameter space. The purpose of designing these combination points is to obtain reflectivity data under different postures as comprehensively as possible within a limited number of tests, providing sufficient samples for subsequent model fitting. "Applying corresponding pressure and angle through the device" refers to the need for precise control of the relative position and force between the device and the standard sample during the calibration process, so that it reaches the preset pressure and angle combination point. This can be achieved in various ways. For example, a high-precision three-dimensional motion platform can be used in conjunction with force and angle sensors to achieve automated and precise control of the device's attitude and pressure; or, a special fixture with a scale and feedback mechanism can be used, allowing the operator to manually adjust and monitor the readings of the pressure sensor 301 and angle sensor 302 in real time to achieve the target attitude. "Multiple measurements of the original reflectivity Roriginal at multiple wavelengths and averaging" refers to illuminating the standard sample sequentially with multiple preset wavelengths using the device's light source 202 at each pressure-angle combination point, and collecting the reflected light signal through the photoelectric detection array 201, repeating the measurement multiple times. Then, the original reflectivity data obtained from each measurement is averaged. The purpose of multiple measurements is to reduce the impact of random noise and instantaneous fluctuations on the measurement results and improve the reliability and accuracy of the data. The original reflectivity Roriginal is a direct measurement value without any attitude correction. "Recording the corresponding pressure P and angle θ to form a calibration dataset" means storing the average raw reflectance Rraw obtained from each measurement, along with the corresponding real-time readings P and θ from pressure sensor 301 and angle sensor 302, together to form a set containing a triple (P, θ, Rraw). This set is the calibration dataset used to construct the correction model. This dataset is the basis for subsequent model fitting, and its quality directly affects the accuracy of the correction model.

[0086] "Standard pressure and standard angle conditions" refer to ideal or expected contact postures, such as the device being in perfect perpendicular contact with the sample surface and applying moderate pressure (e.g., zero tilt angle and a specific standard pressure). These conditions are defined as baseline states, under which reflectance measurements will serve as a reference for measurements under all other postures. "Measured baseline reflectance Rbaseline" refers to taking the average of multi-wavelength reflectance measurements on a standard sample under the aforementioned standard pressure and standard angle conditions, using the same method as measuring the original reflectance. This Rbaseline represents the true reflectance of the standard sample under ideal contact postures and is the gold standard for evaluating the effectiveness of the correction model. "Fitting the deviations at each combination point using polynomial regression" refers to using statistical methods to analyze the difference between the original reflectance Roriginal and the baseline reflectance Rbaseline measured at each combination point in the calibration dataset and establishing a mathematical relationship between this difference and pressure P and angle θ. Polynomial regression is a commonly used curve fitting technique that approximates this nonlinear relationship by constructing polynomial functions of P and θ. Besides multinomial regression, other regression methods, such as support vector regression, Gaussian process regression, or neural network-based regression models, can be used to accommodate more complex nonlinear relationships. "Obtaining a simplified correction formula" refers to a mathematical expression obtained through regression fitting that adjusts the original reflectivity Roriginal according to real-time pressure P and angle θ, thereby approximating the reference reflectivity Rreference. This formula is usually concise and easy to calculate in real-time on a microprocessor. For example, the formula can be expressed as Rcorrected = Roriginal × f(P, θ), where f(P, θ) is a function of P and θ used to compensate for attitude-related deviations.

[0087] "Using cross-validation or splitting the validation set" means that after model training, to objectively evaluate the model's generalization ability and accuracy, it is necessary to test using a dataset independent of the training data. Cross-validation (such as K-fold cross-validation) divides the entire calibration dataset into K parts, using K-1 parts to train the model in turn, and using the remaining part to validate the model. Splitting the validation set divides the dataset into training and validation sets, using the training set to train the model and the validation set to evaluate it. These methods can effectively avoid the overfitting problem where the model performs well on the training data but poorly on new data. "Evaluating the error between the corrected reflectance and the baseline reflectance" means calculating the corrected reflectance Rcorrected from the original reflectance Roriginal in the validation set using the correction formula, and then comparing the corrected Rcorrected with the corresponding baseline reflectance Rbaseline to calculate the difference between the two. Commonly used error metrics include root mean square error (RMSE), mean absolute error (MAE), or maximum relative error. "When the error meets the preset metric" means that during the model evaluation process, if the calculated error value is lower than or equal to a preset threshold, the corrected model is considered to have met the expected accuracy requirements. This preset index can be determined based on the measurement accuracy requirements of the actual application scenario. For example, it can be set to ensure that the relative error between the corrected reflectivity and the reference reflectivity is less than a certain percentage. "Stabilizing the corrected model parameters in the microprocessor" means that once the corrected model passes evaluation and meets the accuracy requirements, its internal coefficients (e.g., coefficients of polynomial regression) will be permanently written into the non-volatile memory (such as flash memory or EEPROM) of the microprocessor in the processing unit 401. The purpose of this stabilization is to ensure that the model parameters are not lost after the device is powered off and can be immediately retrieved after the device is started, thereby achieving real-time and efficient attitude correction functionality.

[0088] In actual detection, "the microprocessor calculates P and θ based on the real-time output of pressure sensor 301 and angle sensor 302" means that during actual wound detection, the microprocessor in processing unit 401 continuously monitors the output signals of pressure sensor 301 and angle sensor 302 connected to it. These sensors provide real-time information on the force (P) and tilt angle (θ) of the user's pressing device. The microprocessor converts these analog or digital signals into physical quantities that can be used to correct the formula. "and calls the above correction formula to compensate for the original reflectivity" means that after obtaining the real-time P and θ values, the microprocessor reads the previously calibrated correction formula and its parameters from its fixed memory, and then substitutes the currently measured multi-wavelength original reflectivity Roriginal, real-time P value, and θ value into the formula for calculation. "Obtaining the target reflectivity Rcorrected after attitude correction" refers to the reflectivity value obtained after calculation by the correction formula. This Rcorrected has eliminated the measurement error introduced by changes in user pressure or angle, making it equivalent to the reflectivity measured under standard posture. This corrected reflectivity R correction will serve as an accurate input for subsequent wound healing quantification level output and healing time prediction.

[0089] Step C: Output of wound healing quantification level; The modified multi-wavelength reflectance is used as input and fed into the wound quantification correlation model trained with clinical annotation data. The corresponding wound healing level (e.g., level 1-10) is output and corresponding text or graphic prompts are given on the display interface. The healing level is divided into multiple levels according to preset standards to characterize different healing stages from superficial repair, granulation growth, deep healing to severe injury. In step C, the wound healing status L is marked as grade 1–10: Grade 1–2: epidermal repair stage; Grade 3–5: granulation tissue growth stage; Grade 6–8: deep healing stage; Grade 9–10: severe injury or significantly delayed healing; For the data samples, the corrected reflectance was measured at multiple wavelengths, and the corresponding clinical healing levels were recorded. When establishing the wound healing level-multi-wavelength reflectance correlation model, the attitude-corrected multi-wavelength reflectance was used as a feature vector: Rcorrection = [Rcorrection(λ1), Rcorrection(λ2), ..., Rcorrection(λn)] The model takes the ratio / difference features derived from the wound healing level as input and the wound healing level L labeled by clinicians as supervision label. It is trained using supervised learning algorithms such as support vector machine, multi-class logistic regression, and neural network to obtain a nonlinear mapping model for realizing L=f(R correction). During online detection, the current multi-wavelength corrected reflectance is input into the model, and the corresponding wound healing level is output.

[0090] Step D: Healing time prediction and risk assessment; Based on historical test data and / or individual physiological characteristics, call the pre-stored healing time prediction model to output the remaining wound healing time and / or healing delay risk warning; In step D, it is determined whether historical detection data for the wound exists in the storage module: When there is no historical data, the AI ​​processing module calls the single prediction mode based on the current healing level, wound type and individual physiological characteristics parameters, and outputs the estimated wound healing cycle or remaining healing days through a deep learning model. When historical data exists, the AI ​​processing module takes the reflectivity sequence and healing level sequence obtained from multiple detections as time series input, calls the trend prediction mode, uses a deep learning model to learn the reflectivity change trend and healing speed, outputs a more refined prediction result of the remaining healing time, and combines the model judgment to give early warning information such as healing delay and infection risk.

[0091] Step E: Result Display and Data Storage; The device displays the current healing level, estimated remaining healing days, and risk warning information on the display unit. At the same time, it stores the corrected reflectivity, attitude parameters, healing level, and prediction results of this test into the local memory for subsequent continuous monitoring and incremental model optimization. If synchronization with the hospital or mobile device is required, the data can be uploaded to an external terminal through the communication module.

[0092] In step D, the relevant AI healing time prediction model is trained and deployed. 1. Data Collection and Labeling In this embodiment, to achieve intelligent prediction of the remaining healing time, long-term follow-up data from several patients were collected, for example: Approximately 800 wound samples were collected; for each sample, multi-wavelength corrected reflectance was recorded daily or periodically during the healing process to form a reflectance time series; the corresponding healing grade series were obtained by a quantitative model; the basic physiological characteristics of the patients, such as age, blood sugar level, and whether they had underlying diseases, were recorded; the total time required for actual healing or the remaining healing days were recorded.

[0093] The dataset above is divided into training, validation, and test sets, for example, in a ratio of approximately 70% : 15% : 15%.

[0094] 2. Construction of LSTM–CNN Fusion Model This embodiment employs a deep learning model that integrates LSTM and CNN as the healing time prediction model. Its structure includes, for example, the following: LSTM layer: the input dimension is the time series of multi-wavelength reflectance and / or healing level, with, for example, 64 hidden nodes, used to learn the time-dependent features of reflectance changes during the healing process; CNN layer: convolutions the multi-wavelength reflectance in a two-dimensional "wavelength-time" space, with, for example, 3×3 kernels, used to extract spatial correlation features between multiple wavelengths; Feature fusion layer: concatenates the temporal features output by LSTM and the spatial features output by CNN, and inputs them along with the patient's physiological features into a fully connected layer; Output layer: outputs a regression value for the remaining healing days or healing completion time, and can also output the classification results of healing delay risk.

[0095] During training, different weights can be set according to features such as blood glucose levels. For example, when blood glucose is higher than a certain threshold, weight adjustments can be introduced through model structure or loss function to reflect the adverse effects of high blood glucose on healing speed.

[0096] 3. Lightweight Model and Embedded Deployment After the model is trained and validated on the test set, it is quantized and pruned using tools such as TensorFlow Lite to compress the number of parameters and storage footprint to a suitable size, such as less than 500 KB. The lightweight model is then embedded into an ARM Cortex-M4 microprocessor or a compatible memory chip to ensure that the single inference time is within hundreds of milliseconds, meeting the real-time requirements of portable devices.

[0097] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications, improvements, and combinations made by those skilled in the art under the guidance of the present invention, such as changing the number and arrangement of OLEDs and OPDs, using different types of pressure / angle sensors, replacing them with other embedded AI chips, or adjusting the deep learning model structure, should all be considered to fall within the protection scope of the present invention.

Claims

1. A device for detecting wound healing status and predicting healing time based on optical reflectivity, characterized in that, include: A flexible substrate (101) has a photoelectric detection area (102) and a light source area (103) arranged around the photoelectric detection area. A photoelectric detection array (201) is disposed within the photoelectric detection area (102), the photoelectric detection array comprising at least one organic photoelectric detector; Multiple light sources (202) are arranged in the light source area (103) and spaced apart in the circumference. The light-emitting surface of each light source faces the photoelectric detection area (102) and is used to irradiate the wound tissue located above the device in sequence with multiple preset wavelengths. A positioning area (104) is disposed on the upper surface of the flexible substrate (101), the planar projection of the positioning area on the flexible substrate at least partially overlaps with the photoelectric detection area (102), for guiding the wound area of ​​the finger or limb to align with the photoelectric detection array (201). At least one pressure sensor (301) is disposed in the positioning area (104) for detecting the contact pressure between the finger and the device; An angle sensor (302) disposed on the flexible substrate (101) and / or the circuit board below it is used to detect the spatial posture of the device or finger pressing direction; The processing unit (401) includes a microprocessor that integrates an analog-to-digital converter, a storage unit and an embedded computing unit. The processing unit is electrically connected to the photoelectric detection array (201), the pressure sensor (301), the angle sensor (302) and the light source (202), respectively.

2. The apparatus as claimed in claim 1, characterized in that, The flexible substrate (101) is a polyimide substrate, covered with a medical-grade elastic encapsulation layer (405). The encapsulation layer forms the positioning area (104) on its upper surface for direct contact with the skin and to provide flexible adhesion.

3. The apparatus as described in claim 1 or 2, characterized in that, The light source has a preset tilt angle relative to the flexible substrate to form annular multi-wavelength oblique incident illumination.

4. The apparatus as claimed in claim 1, characterized in that, The flexible substrate has a photoelectric detection area with a groove at the center of the flexible substrate, and the flexible substrate has a light source area with an annular groove surrounding the outside of the photoelectric detection area.

5. The apparatus as claimed in claim 1, characterized in that, The light source is uniformly arranged circumferentially along the annular light source area, and the light source includes at least four sets of OLED light sources with emission wavelengths covering approximately 450–950 nm.

6. A prediction method based on the wound healing status detection and healing time prediction device according to any one of claims 1 to 5, wherein the device is a portable detection device equipped with an AI processing module, comprising the following steps: Step A: Posture guidance and contact detection; The user places the injured finger or other wound on the positioning area of ​​the device. The pressure sensor detects whether the contact pressure is within the preset range, and the angle sensor detects whether the pressing angle is within the preset range. When the pressure and angle meet the set conditions, a contact status compliance prompt is output, allowing the process to proceed to the optical detection stage. Otherwise, the user is prompted to adjust the pressing posture through an audio-visual signal. Step B: Multi-wavelength optical illumination and reflection signal acquisition; control the OLED light source (202) of each wavelength to emit light in a preset order, and acquire the multi-wavelength reflected photoelectric signals output by the photoelectric detection array (201) and the corresponding pressure and angle information; according to the signal processing and the pre-stored "pressure-angle-reflectivity" correction model, correct the original reflectivity obtained under different pressures and angles to the corrected reflectivity equivalent to the standard posture. Step C: Outputting the quantitative level of wound healing; The corrected multi-wavelength reflectance is used as input and fed into the wound quantitative correlation model trained with clinical annotation data to output the corresponding wound healing level; The healing level is divided into multiple levels according to preset standards to characterize different healing stages from superficial repair, granulation tissue growth, deep healing to severe injury. Step D: Healing time estimation and risk assessment; Based on historical testing data and / or individual physiological characteristics, a pre-stored healing time prediction model is invoked to output the remaining wound healing time and / or a warning of healing delay risk. Step E: Result Display and Data Storage; The current wound healing level, estimated healing time, and risk warnings are visualized and output through the display module or external terminal. At the same time, the multi-wavelength reflectivity, attitude parameters, and prediction results of this test are stored locally or in the cloud for subsequent continuous monitoring and incremental model optimization.

7. The prediction method based on a wound healing status detection and healing time prediction device as described in claim 6, characterized in that, In step D, it is determined whether historical detection data for the wound exists in the storage module: When there is no historical data, the AI ​​processing module calls the single prediction mode based on the current healing level, wound type and individual physiological characteristic parameters, and outputs the estimated wound healing cycle or remaining healing days through the deep learning model. When historical data exists, the AI ​​processing module takes the reflectivity sequence and healing level sequence obtained from multiple detections as time series input, calls the trend prediction mode, uses a deep learning model to learn the reflectivity change trend and healing speed, outputs a more refined prediction result of the remaining healing time, and combines the model judgment to give early warning information such as healing delay and infection risk.

8. The prediction method based on a wound healing status detection and healing time prediction device according to claim 6, characterized in that, The specific steps for constructing the "pressure-angle-reflectivity" correction model are as follows: A standard sample with a smooth, polished quartz sheet and stable reflectivity was selected as the calibration object and fixed in the detection area of ​​the device. Within a preset pressure and angle range, several pressure-angle combination points are designed. At each combination point, the corresponding pressure and angle are applied through the device, and the original reflectivity R of multiple wavelengths is measured multiple times. 原始， Take the average value and record the corresponding pressure P and angle θ to form a calibration dataset; The reference reflectance R was measured under standard pressure and standard angle conditions. 基准 By fitting the deviations of each combination point through polynomial regression, a simplified correction formula is obtained, for example: R 修正 =R 原始 ×(0.98+0.01P-0.002θ) Where P is the pressing force (in N), and θ is the pressing angle (in °). The error between the corrected reflectance and the reference reflectance is evaluated by cross-validation or partitioning the validation set. When the error meets the preset index, the parameters of the corrected model are stored in the microprocessor. In actual testing, the microprocessor calculates P and θ based on the real-time outputs of the pressure and angle sensors, and then uses the aforementioned correction formula to compensate for the original reflectivity, obtaining the attitude-corrected target reflectivity R. 修正 .

9. The prediction method based on a wound healing status detection and healing time prediction device according to claim 6, characterized in that, In step C, the wound healing status L is marked as 1–10: 1–2: epidermal repair stage; 3–5: granulation tissue growth stage; 6–8: deep healing stage; 9–10: severe injury or significantly delayed healing. For the data samples, the corrected reflectance was measured at multiple wavelengths, and the corresponding clinical healing levels were recorded. When establishing the wound healing level-multi-wavelength reflectance correlation model, the attitude-corrected multi-wavelength reflectance was used as a feature vector: R Correction = [Rcorrection(λ1), Rcorrection(λ2), ..., Rcorrection(λn)] The model inputs are the ratio / difference features derived from the wound healing level L labeled by clinicians, and the model is trained using support vector machine / multi-class logistic regression / neural network supervised learning algorithms to obtain a nonlinear mapping model for realizing L=f(R correction). During online detection, the current multi-wavelength corrected reflectance is input into the model, and the corresponding wound healing level is output.

10. The prediction method based on a wound healing status detection and healing time prediction device according to claim 6, characterized in that, The healing time prediction model is a deep learning model that combines LSTM and CNN, and its structure includes: LSTM layer: The input dimension is the time series of multi-wavelength reflectance and / or healing level, used to learn the time-dependent features of reflectance changes during the healing process; CNN layer: Convolves the reflectance of multiple wavelengths in a two-dimensional space of "wavelength-time" to extract spatial correlation features between multiple wavelengths; Feature fusion layer: The temporal features output by LSTM and the spatial features output by CNN are concatenated and input into the fully connected layer along with the patient's physiological features; Output layer: The output is a regression value of the remaining days of healing or the time to complete healing, and can also output the classification results of the risk of delayed healing.

Images