A method and system for monitoring neonatal jaundice without innovation

Abstract

The present application relates to the technical field of non-invasive medical monitoring of newborns, in particular, the present application relates to a non-invasive jaundice monitoring method and system for newborns, in the present application, the light reflection measurement unit adopts a dual-wavelength alternate emission and tilt angle light source design, suppresses the fetal grease mirror reflection, converts the dual-wavelength reflection light intensity ratio into the original transcutaneous bilirubin value, the fetal grease analysis unit accurately identifies the fetal grease area and calculates the coverage through multispectral imaging and convolutional neural network, the dynamic pressure sensing unit monitors the contact pressure in real time, triggers the audible light warning when the value is out of the standard interval, the intelligent calibration processing unit calls the three-dimensional lookup table, matches the correction coefficient through bilinear interpolation, superimposes the gestational age and measurement site compensation factor, and outputs the corrected transcutaneous bilirubin value, eliminates the interference of fetal grease reflection and pressure fluctuation, realizes accurate non-invasive monitoring, and provides a reliable basis for clinical diagnosis of neonatal jaundice.

Description

Technical Field

[0001] This invention relates to the field of non-invasive neonatal medical monitoring technology, and more specifically, to a non-invasive neonatal jaundice monitoring method and system. Background Technology

[0002] Non-invasive neonatal medical monitoring technology is an important technology, specifically applied to the non-invasive monitoring of transcutaneous bilirubin in neonatal jaundice. Its core is to accurately measure bilirubin concentration by precisely handling interference factors on the skin surface and the effects of operational fluctuations, thus meeting the core requirements of non-invasive, accurate and safe neonatal clinical monitoring.

[0003] The natural presence of vernix caseosa on the surface of a newborn's skin alters the light reflection path due to its smooth lipid properties. Additionally, the contact pressure between the probe and the skin can fluctuate due to variations in the operator's technique. These two factors combined lead to distortion of the raw signal from transcutaneous bilirubin measurement, resulting in discrepancies between the monitoring results and the actual serum bilirubin value. Consequently, this method cannot provide a reliable basis for clinical jaundice diagnosis and intervention. To address this technical issue, we have developed a non-invasive neonatal jaundice monitoring method and system. Summary of the Invention

[0004] The purpose of this invention is to provide a non-invasive method and system for monitoring neonatal jaundice, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, one objective of this invention is to provide a non-invasive method for monitoring neonatal jaundice, comprising the following steps:

[0006] S1: Use a transcutaneous bilirubin meter to perform preliminary light reflectance measurements at the newborn's skin measurement points to obtain the raw transcutaneous bilirubin value;

[0007] S2: The skin surface image of the newborn skin measurement point is captured by the miniature optical imaging unit integrated in the probe of the transcutaneous bilirubin meter, and the skin surface image is analyzed by the built-in image processing algorithm to identify the boundary between the vernix caseosa area and the non-vernix caseosa area, obtain the vernix caseosa coverage area, calculate the percentage ratio of the vernix caseosa coverage area to the total area of ​​the newborn skin measurement point as the vernix caseosa coverage value, and obtain the contact pressure value when contacting the newborn skin through the pressure sensor embedded in the probe.

[0008] S3: The calibration model pre-stored in the transcutaneous bilirubinometer is invoked, and the original transcutaneous bilirubin value is calibrated based on the vernix caseosa coverage value and contact pressure value. The calibration model is a lookup table established using experimental data, which maps correction coefficients for different combinations of vernix caseosa coverage percentage and contact pressure value. The calibration process includes:

[0009] Based on the vernix caseosa coverage value and the contact pressure value, the corresponding correction coefficient is matched in the lookup table, and then the correction coefficient is applied to the original transcutaneous bilirubin value. The corrected transcutaneous bilirubin value is calculated and output through linear superposition interpolation to eliminate the signal attenuation caused by vernix caseosa reflection interference and contact pressure fluctuations.

[0010] A second objective of this invention is to provide a system for implementing a non-invasive neonatal jaundice monitoring method as described in any one of the above-mentioned claims, comprising:

[0011] The light reflection measurement unit integrates a dual-wavelength light source module and a photoelectric sensor to execute a dual-wavelength red and blue light alternating emission mode. The light source emitters are arranged at an angle to suppress vernix caseosa specular reflection. The reflectivity calculation module converts the ratio of the intensity of the received dual-wavelength reflected light into the original transdermal bilirubin value, providing initial input for calibration.

[0012] The vernix caseosa analysis unit consists of a miniature optical imaging unit and an image processing algorithm, including a multispectral ring light source, a high-resolution miniature camera, and a convolutional neural network processor. It is used to capture images of the skin surface in real time, identify the boundaries of the vernix caseosa region, calculate the percentage of vernix caseosa coverage, and output the quantified vernix caseosa coverage value to the calibration unit.

[0013] The dynamic pressure sensing unit is embedded in the piezoresistive sensor array on the probe contact surface to monitor the contact pressure value in real time. It integrates a pressure feedback mechanism, which triggers an audible and visual warning and pauses calibration when the pressure deviates from the standard range. The subsequent process is activated after the pressure stabilizes.

[0014] The intelligent calibration processing unit has a built-in microprocessor and memory, storing a three-dimensional lookup table and a compensation factor database. It is used to call up vernix caseosa coverage value, contact pressure value and original transcutaneous bilirubin value, match the correction coefficient through bilinear interpolation and superimpose the compensation factor, and output the corrected transcutaneous bilirubin value to eliminate vernix caseosa interference and pressure fluctuation.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] This invention employs a dual-wavelength alternating emission and tilted light source design in the light reflection measurement unit to suppress interference from vernix caseosa specular reflection. Combined with multispectral imaging and convolutional neural network segmentation in the vernix caseosa analysis unit, it accurately identifies vernix caseosa regions and calculates coverage, thus solving the problem of vernix caseosa altering the light reflection path.

[0017] The dynamic pressure sensing unit monitors the contact pressure in real time. When the pressure exceeds the standard range, it triggers audible and visual feedback to guide the adjustment to a stable range, eliminating the influence of pressure fluctuations caused by differences in operating techniques. The intelligent calibration unit calls a three-dimensional lookup table, matches the correction coefficient through bilinear interpolation, and superimposes gestational age and measurement site compensation factors to accurately calibrate the distortion of the original signal. The final output corrected percutaneous bilirubin value is close to the real serum value, achieving non-invasive and accurate monitoring and providing a reliable basis for the clinical diagnosis and intervention of neonatal jaundice. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall workflow of the present invention;

[0019] Figure 2 This is a schematic diagram of the overall structure of the present invention;

[0020] The meanings of the labels in the diagram are as follows:

[0021] 1. Light reflection measurement unit; 2. Vernix caseosa analysis unit; 3. Dynamic pressure sensing unit; 4. Intelligent calibration processing unit. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 As shown, one of the objectives of this embodiment is to provide a non-invasive method for monitoring neonatal jaundice, including the following steps:

[0024] S1: Use a transcutaneous bilirubin meter to perform preliminary light reflectance measurements at the newborn's skin measurement points to obtain the raw transcutaneous bilirubin value;

[0025] S2: The micro-optical imaging unit integrated into the probe of the transcutaneous bilirubinometer captures the skin surface image of the newborn skin measurement point, and uses the built-in image processing algorithm to analyze the skin surface image, identify the boundary between the vernix caseosa area and the non-vernix caseosa area, obtain the vernix caseosa coverage area, calculate the percentage ratio of the vernix caseosa coverage area to the total area of ​​the newborn skin measurement point as the vernix caseosa coverage value, and obtain the contact pressure value when contacting the newborn skin through the pressure sensor embedded in the probe.

[0026] S3: Call the calibration model pre-stored in the transcutaneous bilirubinometer and calibrate the original transcutaneous bilirubin value based on the vernix caseosa coverage value and contact pressure value. The calibration model is a lookup table established through experimental data. The lookup table maps the correction coefficients under different combinations of vernix caseosa coverage percentage and contact pressure value. The calibration process includes:

[0027] Based on the vernix caseosa coverage value and contact pressure value, the corresponding correction coefficient is matched in the lookup table. Then, the correction coefficient is applied to the original transcutaneous bilirubin value, and the corrected transcutaneous bilirubin value is calculated and output through linear superposition interpolation to eliminate the signal attenuation caused by vernix caseosa reflection interference and contact pressure fluctuations.

[0028] When performing preliminary light reflectance measurements at the skin measurement points of newborns using a transcutaneous bilirubinometer, a dual-wavelength alternating emission mode of red and blue light is adopted. The red light wavelength range is the first specific band, and the blue light wavelength range is the second specific band. The photoelectric sensor inside the probe simultaneously receives the intensity of reflected light of the two wavelengths, and the ratio of the intensity of reflected light of the two wavelengths is converted into the original transcutaneous bilirubin value through the built-in reflectivity calculation module. The light source emitting end of the probe is designed at an angle to the skin measurement surface to avoid direct light forming specular reflection interference on the vernix caseosa surface.

[0029] The miniature optical imaging unit includes a multispectral ring light source and a high-resolution miniature camera. The multispectral ring light source is arranged around the probe measurement window and simultaneously emits a combination of visible light and near-infrared band illumination during light reflection measurement. The miniature camera captures images of the skin surface from a vertical perspective. The image resolution satisfies the requirement that the actual skin size corresponding to a single pixel is smaller than the minimum feature size of the vernix caseosa. The built-in image processing algorithm first performs distortion correction and illumination equalization preprocessing on the image, and then segments the vernix caseosa region through a convolutional neural network. This network uses the texture features and reflective properties of vernix caseosa as training labels.

[0030] When identifying the boundary between the vernix caseosa region and the non-vernix caseosa region, based on the binary segmentation map output by the convolutional neural network, morphological closing operations are used to fill the tiny pores in the vernix caseosa sheet. Then, continuous closed boundaries are extracted by the edge detection algorithm. The vernix caseosa coverage area is obtained by counting the total number of pixels within the closed boundaries. At the same time, isolated areas with an area smaller than the vernix caseosa fragment threshold are removed to eliminate measurement noise.

[0031] When calculating the percentage of vernix caseosa coverage area to the total area of ​​measurement points, the total area of ​​measurement points is defined as the projection area of ​​the probe measurement window on the skin surface. The projection area is defined by the bounding rectangle of the closed boundary. The vernix caseosa coverage value is calculated as the percentage of the number of vernix caseosa pixels divided by the total number of pixels in the projection area. The calculation result is rounded to the nearest integer percentage, and edge pixel distortion caused by probe tilt is compensated by interpolation.

[0032] When calling the pre-stored calibration model, the microprocessor retrieves the three-dimensional lookup table in real time by combining the original transcutaneous bilirubin value, vernix caseosa coverage value and contact pressure value. The calibration model is activated when the probe contact pressure is within the standard pressure range. If the pressure sensor detects that the contact pressure exceeds the standard pressure range, the pressure feedback mechanism is triggered first to require the operator to adjust the pressure. The calibration process is then executed after the pressure stabilizes within the standard range.

[0033] The lookup table was constructed through clinical trials, with experimental samples covering the forehead, chest, and limbs of preterm and full-term infants. Raw percutaneous bilirubin values ​​were collected under a combination of vernix caseosa coverage gradients and contact pressure gradients, and regression analysis was performed with the true serum bilirubin values ​​obtained from simultaneous venous blood tests. The correction coefficient for each data unit was defined as the average deviation between the true serum bilirubin value and the raw percutaneous bilirubin value. The lookup table is stored as a three-dimensional matrix with vernix caseosa coverage percentage as rows, contact pressure values ​​as columns, and correction coefficients as values.

[0034] When matching the correction coefficient in the lookup table, the four nearest neighbor data cells are located using the vernix caseosa coverage value as the row index and the contact pressure value as the column index. If the vernix caseosa coverage value or the contact pressure value is between the two indices, the actual correction coefficient is calculated using bilinear interpolation. When applying the correction coefficient to the original transcutaneous bilirubin value, it is directly algebraically added to the original transcutaneous bilirubin value to generate an intermediate calibration value.

[0035] When calculating and outputting the corrected transcutaneous bilirubin value using linear superposition interpolation, a compensation factor related to the newborn's gestational age and measurement site is further introduced for the intermediate calibration value. The compensation factor is pre-stored in the calibration model and its weight is automatically matched according to the input newborn's gestational age and measurement point location. The final corrected value is output as a weighted sum of the intermediate calibration value and the compensation factor to eliminate residual errors caused by individual physiological differences.

[0036] Further explanation is needed regarding the initial light reflectance measurement for non-invasive neonatal jaundice monitoring. To accurately capture the light reflectance signal that reflects bilirubin concentration while minimizing interference from vernix caseosa on the newborn's skin, the transcutaneous bilirubinometer employs a combination of alternating dual-wavelength emission and a tilted light source design. The specific implementation method is as follows:

[0037] When using a transcutaneous bilirubinometer to perform preliminary light reflectance measurements at measurement points on a newborn's skin, the core technology employs a dual-wavelength alternating red and blue light emission mode. This mode is achieved through a dual-wavelength light source module integrated within the probe. The red light wavelength range is set as the first specific band, which has specific absorption characteristics for bilirubin and can effectively reflect differences in bilirubin concentration. The blue light wavelength range is set as the second specific band, which serves as a reference wavelength and is less affected by bilirubin concentration, thus offsetting interference from irrelevant factors such as skin thickness and pigmentation. The light source module contains two independent light-emitting diodes, corresponding to the red and blue light bands respectively. Alternating emission is achieved through a timing control circuit, with the emission frequency set to 100Hz to ensure that the light reflection signals of the two wavelengths do not overlap and can be rapidly and continuously acquired. The photoelectric sensor inside the probe simultaneously receives the reflections of both wavelengths. For light intensity, the sensor uses a high-sensitivity silicon photodiode with a spectral response range covering 400-1100nm. It can accurately capture reflected light signals of red and blue light, converting the light signals into corresponding electrical signals. After being amplified by a preamplifier circuit, the electrical signals are transmitted to the built-in reflectivity calculation module. The module first converts the current signal into a voltage signal, and then calculates the ratio of the reflected light intensity of red light to that of blue light. Since the absorption of red light by bilirubin increases with increasing concentration, the reflected light intensity of red light decreases, causing the ratio to fluctuate regularly with changes in bilirubin concentration. The reflectivity calculation module pre-stores the intensity ratio and bilirubin concentration mapping curve calibrated in clinical trials. Through this curve, the ratio of the reflected light intensity of the two wavelengths is directly converted into the original transdermal bilirubin value. The conversion process is completed in real time with a delay of no more than 10ms, ensuring measurement efficiency.

[0038] To avoid interference from specular reflection caused by direct light on the vernix caseosa surface, which is a smooth lipid film, vertically incident light is prone to specular reflection, leading to distortion of the reflected light signal received by the sensor. The light source emitting end of the probe is designed at an angle to the skin measurement surface. The size of the angle was determined to be 30-45 degrees through experimental calibration. This angle can ensure that the light effectively illuminates the skin measurement point and produces diffuse reflection, while also deflecting the specular reflection light from the vernix caseosa surface away from the receiving direction of the photoelectric sensor. This ensures that the sensor mainly receives the effective light signal after diffuse reflection by the skin tissue. The outer shell of the light source emitting end adopts a light-shielding design to further prevent stray light from entering the measurement area, ensuring the accuracy of the reflected light intensity measurement and providing reliable raw data input for subsequent calibration.

[0039] While acquiring raw transcutaneous bilirubin values, to accurately analyze the vernix caseosa coverage at measurement points on the newborn's skin, the transcutaneous bilirubinometer probe integrates a miniature optical imaging unit. Through multispectral illumination, high-resolution imaging, and intelligent image processing algorithms, it achieves accurate identification and segmentation of the vernix caseosa region. The specific implementation method is as follows:

[0040] The core of the miniature optical imaging unit comprises a multispectral ring light source and a high-resolution miniature camera, both integrated at the probe's front end and working collaboratively around the measurement window. The multispectral ring light source, composed of eight independent LED light-emitting units, is evenly distributed around the probe's measurement window. During light reflection measurement, it simultaneously emits a combination of visible and near-infrared illumination. The visible light band (400-760nm) clearly reveals the texture details of the skin surface, while the near-infrared band (760-1000nm) penetrates superficial skin tissue, highlighting the difference in reflectivity between vernix caseosa and other skin tissue. The two wavelengths are emitted in a 1:1 intensity ratio, forming a uniform ring illumination field through a diffuser lens, avoiding image distortion caused by localized shadows. The high-resolution miniature camera... The head uses a CMOS image sensor to capture images of the skin surface from a vertical viewing angle (90 degrees to the skin measurement surface). This viewing angle design minimizes perspective distortion and ensures that the pixel size of each area in the image is consistent with the actual skin size. The camera's image resolution is set to 1280×720 pixels. Through optical lens focal length calibration, the single pixel corresponds to an actual skin size of 10μm, while the minimum feature size of vernix caseosa is about 20μm. This resolution can completely capture the fine structure of vernix caseosa and avoid missing vernix caseosa areas due to low pixel count. The camera's frame rate is set to 30fps, capturing 3 frames per measurement. Subsequent average images are taken between frames for processing to eliminate image blurring caused by slight movements of the newborn during the measurement process.

[0041] The built-in image processing algorithm follows a preprocessing-then-segmentation workflow. First, the captured image undergoes distortion correction and illumination equalization preprocessing. Distortion correction uses pre-stored camera lens distortion parameters and a polynomial correction algorithm to correct stretching or compression distortion at image edges. Illumination equalization employs an adaptive histogram equalization algorithm to adjust for differences in brightness caused by uneven lighting, enhancing the contrast between vernix caseosa and non-vernix caseosa areas and preventing localized overexposure or underexposure from affecting subsequent segmentation. After preprocessing, the vernix caseosa area is segmented using a convolutional neural network. This neural network is designed for lightweight... The U-Net network, designed for embedded computing power in transcutaneous bilirubinometers, utilizes the texture and reflectivity of vernix caseosa. Texture features highlight the smooth, poreless, and rounded edges of vernix caseosa, while reflectivity indicates spectral differences in multispectral images. The training dataset contains 5000 clinically collected neonatal skin images, manually labeled for network training. The trained network outputs a binary segmentation map where white pixels represent vernix caseosa regions and black pixels represent skin areas without vernix caseosa, providing accurate data for subsequent calculations of vernix caseosa coverage.

[0042] After the convolutional neural network outputs a binary segmentation image where the vernix caseosa region is white and the non-vernix caseosa region is black, the segmentation image may contain tiny pores and isolated noise points within the vernix caseosa flakes. Directly extracting the boundary will lead to boundary discontinuities or area statistical biases. Therefore, it is necessary to optimize the boundary recognition accuracy through a process of morphological processing, edge detection, and noise removal. The specific implementation method is as follows:

[0043] When identifying the boundary between the vernix caseosa region and the non-vernix caseosa region, morphological closing operations are first used to fill the tiny pores within the vernix caseosa sheet based on the binarized segmentation image output by the convolutional neural network. Morphological closing is an image processing operation that first dilates and then erodes. The dilation operation expands the white pixel range of the vernix caseosa region, filling the pores, while the erosion operation shrinks the vernix caseosa region back to its original contour, avoiding boundary expansion caused by excessive dilation. A circular structuring element with a radius of 2 pixels is selected for the operation. This size was determined experimentally to fill tiny pores with a diameter of 4 pixels or less without destroying the true boundary of the vernix caseosa. The structuring element is slid pixel by pixel. The image is convolved with the binary segmentation image using a dynamic method. After the operation, the pore areas are filled with white pixels, and the vernix caseosa area forms a complete connected region. After pore filling, continuous closed boundaries are extracted using an edge detection algorithm. Here, the Canny edge detection algorithm is selected, which can effectively suppress noise and extract fine edges. In specific implementation, the processed binary segmentation image is first Gaussian filtered with a standard deviation of 1.0 to filter out residual small noise. Then, the gradient magnitude and direction of the image are calculated. Subsequently, non-maximum suppression is used to eliminate non-peak pixels on the edges. Finally, dual thresholds are set: a high threshold of 80 and a low threshold of 40. Based on the grayscale distribution characteristics of the skin image, pixels above a high threshold are marked as strong edges, while pixels between the high and low thresholds and connected to strong edges are marked as weak edges. These are then integrated to form a continuous, closed vernix caseosa region boundary, ensuring no breaks or redundancy. The vernix caseosa coverage area is obtained by counting the total number of pixels within the closed boundary. Specifically, this involves traversing the region enclosed by the closed boundary in the binarized segmentation image and counting the number of white pixels. Each white pixel represents one unit area, with units of 10μm × 10μm, matching the camera resolution. The counting process is executed by the image processing unit of the embedded microprocessor, ensuring... To ensure real-time performance and eliminate measurement noise, isolated areas smaller than the vernix caseosa fragment threshold need to be removed. The vernix caseosa fragment threshold is a pixel number threshold set based on the minimum effective size of vernix caseosa in clinical practice. By measuring vernix caseosa fragments on the skin surface of 100 newborns, the minimum effective vernix caseosa fragment was determined to have 50 pixels, i.e., an actual size of 500μm×100μm. Therefore, the threshold was set to 50 pixels. All connected regions in the segmentation map were traversed. If the number of pixels in a certain region was less than 50, it was determined to be a noise region and marked as a black pixel. It was not included in the vernix caseosa coverage area statistics to ensure that the statistical results only reflect the real vernix caseosa area.

[0044] After completing the statistics on the area covered by vernix caseosa, it is necessary to further calculate its percentage of the total area of ​​the measurement points to quantify the degree of vernix caseosa coverage. The definition of the total area of ​​the measurement points and the correction of edge distortion directly affect the accuracy of the percentage calculation. Therefore, it is necessary to define the projection area range and compensate for errors through interpolation. The specific implementation method is as follows:

[0045] When calculating the percentage of vernix caseosa coverage area to the total area of ​​measurement points, the total area of ​​measurement points is first defined as the projection area of ​​the probe measurement window on the skin surface. The probe measurement window is a circular window at the front end of the percutaneous bilirubinometer probe used to contact the skin, and its physical diameter is determined by design to be 10mm. When the probe is in perpendicular contact with the skin, the projection area is a circle with a diameter of 10mm. However, to simplify pixel counting, the projection area is limited by a closed bounding rectangle. That is, by traversing all pixels of the vernix caseosa region in the binarized segmented image, the maximum x-coordinate, minimum x-coordinate, maximum y-coordinate, and minimum y-coordinate of each pixel are obtained. The rectangle formed by these four coordinates is the bounding rectangle. This rectangle completely includes the vernix caseosa region and the probe projection range. The total number of pixels in this rectangle is the number of pixels corresponding to the total area of ​​measurement points. The vernix caseosa coverage value is calculated as the percentage of the number of pixels covered by vernix caseosa divided by the total number of pixels in the projection area, i.e., coverage value = (number of pixels covered by vernix caseosa / total number of pixels in the projection area) × 100%. The calculation process is performed by a microprocessor. The arithmetic logic unit of the device executes the calculation. For example, if the number of vernix caseosa pixels is 300,000 and the total number of pixels in the projection area is 1,000,000, the coverage value is 30%. The calculation result is rounded to the nearest integer percentage to ensure that the value is concise and in line with clinical usage habits. In actual operation, the probe may be tilted (not perpendicular to the skin), which may cause the pixels at the edge of the projection area to be distorted. This distortion needs to be compensated by interpolation. Specifically, the edge distortion area is first determined, which is usually the range of 5 pixels of the outer rectangle edge. Based on the maximum tilt angle of the probe of 30 degrees, this range can cover all possible distorted pixels. Then, bilinear interpolation is used to calculate the equivalent gray value of the distorted pixel based on the gray value of the 4 effective pixels around the distorted pixel (clear pixels in the non-edge area). If the equivalent gray value is greater than or equal to 128 (binarization threshold), it is determined to be a vernix caseosa pixel and included in the coverage area statistics. Otherwise, it is determined to be a pixel without vernix caseosa. This method corrects the counting error of edge pixels and ensures the accuracy of coverage calculation.

[0046] After obtaining the original percutaneous bilirubin value, vernix caseosa coverage value, and contact pressure value, it is necessary to call the pre-stored calibration model to calibrate the original values ​​to eliminate the influence of vernix caseosa and pressure fluctuations. The effective use of the calibration model depends on the contact pressure being within a reasonable range. Therefore, it is necessary to simultaneously perform lookup table retrieval and pressure compliance judgment. The specific implementation method is as follows:

[0047] When calling the pre-stored calibration model, the microprocessor built into the transcutaneous bilirubin meter first combines the original transcutaneous bilirubin value, vernix caseosa coverage value, and contact pressure value to retrieve the three-dimensional lookup table in real time. The three-dimensional lookup table is structured data pre-stored in the instrument's Flash memory. Its data structure uses vernix caseosa coverage percentage as the row index, contact pressure value as the column index, and correction coefficient as the data value, reflecting the deviation between the original value and the true value under this combination of vernix caseosa and pressure. The microprocessor determines the row index through vernix caseosa coverage value and the column index through contact pressure value, quickly locating the corresponding correction coefficient to ensure real-time calibration.

[0048] The calibration model is only activated when the probe contact pressure is within the standard pressure range. This standard pressure range is based on pressure ranges determined through clinical trials. Testing data from 500 neonatal skin measurements revealed that when the contact pressure is 50-100 kPa, the probe maintains close contact with the skin without compressing it and causing tissue deformation. At this pressure, the light reflection signal is stable. Therefore, the standard pressure range is set to 50-100 kPa. This range parameter is pre-stored in the configuration register of the pressure detection module. If the piezoresistive sensor array of the dynamic pressure sensing unit detects that the contact pressure exceeds the standard pressure range, the pressure feedback mechanism is triggered first, requiring the operator to adjust the pressure. The pressure feedback mechanism is activated by the instrument... The instrument's audible and visual warning module enables the following: when the pressure is below 50 kPa, the green LED at the front of the instrument flashes, and the buzzer emits a low-frequency warning sound; when the pressure is above 100 kPa, the red LED flashes, and the buzzer emits a high-frequency warning sound. The operator adjusts the probe pressure according to the prompts. The pressure sensor monitors and provides feedback on the pressure value in real time until the pressure stabilizes in the 50-100 kPa range. The judgment criterion is that five consecutive sampling values ​​are within the range. At this point, the pressure feedback mechanism stops warning, and the microprocessor restarts the calibration process, calling a three-dimensional lookup table to match the correction coefficients, ensuring that the calibration is performed under compliant pressure conditions and avoiding calibration deviations caused by abnormal pressure.

[0049] The three-dimensional lookup table is not derived theoretically, but rather constructed through clinical trials covering multiple scenarios to ensure that the correction coefficients accurately match the actual physiological differences and measurement conditions of newborns. The specific implementation method is as follows:

[0050] When constructing the lookup table through clinical trials, the first step is to determine the sample size. The sample must cover two groups: preterm infants (gestational age 28-36 weeks) and full-term infants (gestational age 37-42 weeks), with 200 cases selected from each group. The sample must also cover different birth weights (preterm infants 1000-2500g, full-term infants 2500-4000g) to avoid bias caused by a single group. Measurement sites are the forehead, chest, and limbs (outer upper arms), common in newborns. These three sites represent typical scenarios of thin skin (forehead), evenly distributed fat (chest), and thicker skin (limbs), respectively. Data must be collected from each site to ensure the lookup table is suitable. For measurement needs at different sites, during the experimental data acquisition phase, raw transcutaneous bilirubin values ​​need to be collected according to a combination of vernix caseosa coverage gradient and contact pressure gradient. The vernix caseosa coverage gradient is a range of vernix caseosa coverage set at fixed intervals. Based on the common degree of vernix caseosa coverage in clinical practice, each gradient is achieved by manually adjusting the vernix caseosa on the newborn's skin surface, and the actual coverage is confirmed by a miniature optical imaging unit. The contact pressure gradient is a range of contact pressure set at fixed intervals, referencing the standard pressure range (50-100 kPa). The probe contact pressure is precisely controlled by the pressure adjustment module of the transcutaneous bilirubin meter, and the pressure value is fed back in real time by the dynamic pressure sensing unit.

[0051] For each combination of vernix caseosa coverage and contact pressure gradients, three light reflectance measurements were performed at selected sites using a transcutaneous bilirubinometer. The average value was taken as the raw transcutaneous bilirubin value for that combination to ensure data repeatability. Simultaneously, a true serum bilirubin value was obtained through venous blood testing, as venous blood testing is the clinically recognized gold standard for bilirubin concentration. The time interval between the collection and transcutaneous measurement should not exceed 5 minutes to avoid deviations caused by changes in bilirubin concentration over time. The vanadate oxidation method was used for testing, performed by professional equipment in the hospital's laboratory. The test result was accurate to 0.1 mg / dL and was taken as the true serum bilirubin value. Subsequently, regression analysis was performed between the raw transcutaneous bilirubin value and the true serum bilirubin value for the same combination. The regression analysis was performed using statistical software to calculate the deviation (true) of each data set. The original value is calculated by first calculating the deviation of all samples under the same combination, and then averaging the deviations. This average value is the correction coefficient corresponding to the combination of vernix caseosa coverage and contact pressure. Finally, all correction coefficients are stored in a three-dimensional matrix to construct a lookup table. The row dimension of the matrix is ​​the percentage of vernix caseosa coverage, and the column dimension is the contact pressure value. Each cell in the matrix is ​​the correction coefficient for the corresponding combination. The storage medium is the Flash memory of the transcutaneous bilirubinometer, and the storage format is binary. The row index and column index are quickly located through offset address. For example, vernix caseosa coverage of 30% corresponds to row index 6 (0% is index 0, incrementing by 1 for every 5%), and contact pressure of 80 kPa corresponds to column index 8 (0 kPa is index 0, incrementing by 1 for every 10 kPa). This enables fast reading of the correction coefficients and lays the data foundation for subsequent real-time calibration.

[0052] After completing the lookup table construction and confirming that the probe contact pressure is within the standard range, the intelligent calibration processing unit needs to match the corresponding correction coefficient from the lookup table based on the current vernix caseosa coverage value and contact pressure value. If the value is between the indices, it is supplemented by interpolation, and then combined with the original value to generate an intermediate calibration value. The specific implementation method is as follows:

[0053] When matching the correction coefficient in the lookup table, the four nearest neighbor data cells are first located using the vernix caseosa coverage value as the row index and the contact pressure value as the column index. These four nearest neighbor data cells are the four cells enclosed by the two closest row indices on either side of the current vernix caseosa coverage value and the two closest column indices on either side of the current contact pressure value. For example, if the current vernix caseosa coverage value is 32%, the corresponding row indices in the lookup table are 30% (index 6) and 35% (index 7), thus determining the two row indices. Similarly, if the current contact pressure value is 85 kPa, the corresponding column indices are 80 kPa (index 8) and 90 kPa (index 9), thus determining the two column indices. The four data cells are (30%, ... The data is divided into four indexes: (80 kPa), (30%, 90 kPa), (35%, 80 kPa), and (35%, 90 kPa). Each cell stores a corresponding correction coefficient (e.g., 0.6 mg / dL, 0.7 mg / dL, 0.8 mg / dL, and 0.9 mg / dL respectively). If the vernix caseosa coverage value or contact pressure value is between two indices, the actual correction coefficient is calculated using bilinear interpolation. Bilinear interpolation calculates the intermediate value by assigning weights to the correction coefficients of four adjacent data cells according to the distance between the current value and the index value. Specifically, interpolation is first performed in the column direction (pressure dimension). For row index 6 (30%), the distance between 85 kPa and 80 kPa and 90 kPa is calculated. With weights (85 kPa 5 kPa from 80 kPa and 5 kPa from 90 kPa, both weighted at 0.5), the correction factor for 85 kPa at 30% coverage is 0.6 × 0.5 + 0.7 × 0.5 = 0.65 mg / dL. Similarly, for row index 7 (35%), the correction factor for 85 kPa at 35% coverage is calculated as 0.8 × 0.5 + 0.9 × 0.5 = 0.85 mg / dL. Then, interpolating along the row direction (coverage dimension): 32% from 30% (2%) and from 35% (3%), with weights of 0.6 (3 / 5) and 0.4 (2 / 5) respectively, the actual correction factor is 0.65 × 0.6 + 0.85 × 0.4 = 0.73 mg / dL. This value is... The correction coefficient, adapted to the current measurement conditions, is applied to the original transcutaneous bilirubin value. An intermediate calibration value is generated by direct algebraic addition. Since the correction coefficient is defined as the average deviation between the true serum bilirubin value and the original transcutaneous bilirubin value, algebraic addition can directly offset the deviation between the original and true values. For example, if the original transcutaneous bilirubin value is 8.2 mg / dL and the correction coefficient is 0.73 mg / dL, the intermediate calibration value = 8.2 + 0.73 = 8.93 mg / dL. If the correction coefficient is negative, the algebraic addition will lower the value, ensuring the intermediate calibration value is closer to the true value. The addition operation is performed by the arithmetic logic unit of the intelligent calibration processing unit, providing accurate intermediate data for the subsequent introduction of compensation factors.

[0054] Although the intermediate calibration values ​​have eliminated the influence of vernix caseosa and pressure fluctuations, individual physiological differences among newborns may still lead to residual errors. Therefore, it is necessary to introduce a compensation factor through linear superposition interpolation to further optimize the calibration results. The specific implementation method is as follows:

[0055] When calculating the corrected transcutaneous bilirubin value using linear superposition interpolation, a compensation factor related to the neonatal gestational age and measurement site is first introduced for the intermediate calibration value. This compensation factor is a quantitative correction value based on the influence of individual neonatal differences on bilirubin measurement; its magnitude is determined through additional clinical trials. The trials selected 50 neonates each with different gestational ages (28-42 weeks, with each interval being 2 weeks) and different measurement sites (forehead, chest, limbs). Transcutaneous measurements and true venous blood values ​​were collected for each combination of gestational age interval and measurement site. The average residual deviation between the intermediate calibration value and the true value for that combination was calculated; this deviation is the mean relative deviation. The compensation factors for the corresponding combinations reflect the direction and degree of deviation in intermediate calibration values ​​caused by individual differences. The compensation factors are pre-stored in the compensation factor database of the calibration model. The database is stored in a two-dimensional structure according to gestational age range and measurement site. Each combination corresponds to a unique compensation factor. The storage medium and lookup table share the same Flash memory, which can be quickly retrieved through index. For example, if the newborn's gestational age is 32 weeks (corresponding to the gestational age range of 32-34 weeks, index 5) and the measurement site is the forehead (index 0), then the database address is calculated by the formula gestational age index × number of sites + site index, and the compensation factor is quickly matched.

[0056] In practice, before measurement, medical staff need to input the newborn's gestational age and measurement site through the transcutaneous bilirubin meter's interface. The system automatically maps the gestational age to the corresponding gestational age range and then matches the corresponding compensation factor. No manual calculation is required. The final correction value is output as a weighted sum of the intermediate calibration value and the compensation factor. The weighting of the sum is based on the reliability of the compensation factor. Since the compensation factor has been validated by a large amount of clinical data and has high reliability, the intermediate calibration value is weighted at 0.9 and the compensation factor at 0.1. This weighting has been statistically validated and can maximize the offsetting of residual errors without introducing new ones. For example, if the intermediate calibration value is 8.93 mg / dL and the compensation factor is 0.2 mg / dL, the final correction value is 8.93 × 0.9 + 0.2 × 0.1 = 8.037 + 0.02 = 8.057 mg / dL. After being retained to 0.01 mg / dL, it becomes 8.06 mg / dL. This output value has eliminated errors caused by vernix caseosa reflection interference, contact pressure fluctuations, and individual physiological differences. It is directly displayed on the screen of the transcutaneous bilirubin meter and stored in the instrument's measurement log for easy follow-up and analysis, providing accurate non-invasive monitoring data for the clinical diagnosis of neonatal jaundice.

[0057] The second objective of this invention is to provide a system for implementing a non-invasive neonatal jaundice monitoring method including any of the above-mentioned features, comprising:

[0058] The light reflection measurement unit 1 integrates a dual-wavelength light source module and a photoelectric sensor to execute the dual-wavelength red and blue light alternating emission mode. The light source emission ends are arranged at an angle to suppress vernix caseosa specular reflection. The reflectivity calculation module converts the ratio of the intensity of the received dual-wavelength reflected light into the original transdermal bilirubin value, providing the initial input for calibration.

[0059] The vernix caseosa analysis unit 2 consists of a miniature optical imaging unit and an image processing algorithm, including a multispectral ring light source, a high-resolution miniature camera and a convolutional neural network processor, used to capture skin surface images in real time, identify vernix caseosa region boundaries and calculate vernix caseosa coverage percentage, and output the quantified vernix caseosa coverage value to the calibration unit.

[0060] The dynamic pressure sensing unit 3 is embedded in the piezoresistive sensor array on the probe contact surface to monitor the contact pressure value in real time. It integrates a pressure feedback mechanism, which triggers an audible and visual warning and pauses calibration when the pressure deviates from the standard range. It then activates the subsequent process after the pressure stabilizes.

[0061] The intelligent calibration processing unit 4 has a built-in microprocessor and memory, which stores a three-dimensional lookup table and a compensation factor database. It is used to call up the vernix caseosa coverage value, contact pressure value and original transcutaneous bilirubin value, match the correction coefficient through bilinear interpolation and superimpose the compensation factor, and output the corrected transcutaneous bilirubin value to eliminate vernix caseosa interference and pressure fluctuation.

[0062] In this invention, the light reflection measurement unit adopts a dual-wavelength alternating emission and tilted angle light source design to suppress vernix caseosa specular reflection and convert the intensity ratio of the dual-wavelength reflected light into the original transcutaneous bilirubin value. The vernix caseosa analysis unit accurately identifies the vernix caseosa region and calculates the coverage through multispectral imaging and convolutional neural network. The dynamic pressure sensing unit monitors the contact pressure in real time and triggers an audible and visual warning when the pressure exceeds the standard range. The intelligent calibration processing unit calls a three-dimensional lookup table, matches the correction coefficient through bilinear interpolation, and superimposes the gestational age and measurement site compensation factors to output the corrected transcutaneous bilirubin value, eliminating the interference of vernix caseosa reflection and pressure fluctuation, achieving accurate and non-invasive monitoring, and providing a reliable basis for the clinical diagnosis of neonatal jaundice.

[0063] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-invasive method for monitoring neonatal jaundice, characterized in that: Includes the following steps: S1: Use a transcutaneous bilirubin meter to perform preliminary light reflectance measurements at the newborn's skin measurement points to obtain the raw transcutaneous bilirubin value; S2: The skin surface image of the newborn skin measurement point is captured by the miniature optical imaging unit integrated in the probe of the transcutaneous bilirubin meter, and the skin surface image is analyzed by the built-in image processing algorithm to identify the boundary between the vernix caseosa area and the non-vernix caseosa area, obtain the vernix caseosa coverage area, calculate the percentage ratio of the vernix caseosa coverage area to the total area of ​​the newborn skin measurement point as the vernix caseosa coverage value, and obtain the contact pressure value when contacting the newborn skin through the pressure sensor embedded in the probe. S3: The calibration model pre-stored in the transcutaneous bilirubinometer is invoked, and the original transcutaneous bilirubin value is calibrated based on the vernix caseosa coverage value and contact pressure value. The calibration model is a lookup table established using experimental data, which maps correction coefficients for different combinations of vernix caseosa coverage percentage and contact pressure value. The calibration process includes: Based on the vernix caseosa coverage value and contact pressure value, the corresponding correction coefficient is matched in the lookup table, and then the correction coefficient is applied to the original transcutaneous bilirubin value. The corrected transcutaneous bilirubin value is calculated and output through linear superposition interpolation. When calculating the percentage of vernix caseosa coverage area to the total area of ​​measurement points, the total area of ​​measurement points is defined as the projection area of ​​the probe measurement window on the skin surface. The projection area is defined by the bounding rectangle of the closed boundary. The vernix caseosa coverage value is calculated as the percentage of the number of vernix caseosa pixels divided by the total number of pixels in the projection area. The calculation result is rounded to the nearest integer percentage, and edge pixel distortion caused by probe tilt is compensated by interpolation. When calling the pre-stored calibration model, the microprocessor retrieves the three-dimensional lookup table in real time by combining the original transcutaneous bilirubin value, vernix caseosa coverage value and contact pressure value. The calibration model is activated when the probe contact pressure is within the standard pressure range. If the pressure sensor detects that the contact pressure exceeds the standard pressure range, the pressure feedback mechanism is triggered first to require the operator to adjust the pressure. The calibration process is then executed after the pressure stabilizes within the standard range. The lookup table was constructed through clinical trials, with experimental samples covering the forehead, chest, and limbs of preterm and full-term infants. Raw percutaneous bilirubin values ​​were collected under a combination of vernix caseosa coverage gradients and contact pressure gradients, and regression analysis was performed with the true serum bilirubin values ​​obtained from simultaneous venous blood tests. The correction coefficient for each data unit was defined as the average deviation between the true serum bilirubin value and the raw percutaneous bilirubin value. The lookup table is stored as a three-dimensional matrix with vernix caseosa coverage percentage as rows, contact pressure values ​​as columns, and correction coefficients as values.

2. The non-invasive neonatal jaundice monitoring method according to claim 1, characterized in that: When performing preliminary light reflectance measurements at the skin measurement points of newborns using a transcutaneous bilirubinometer, a dual-wavelength red and blue light alternating emission mode is adopted. The photoelectric sensor inside the probe simultaneously receives the intensity of reflected light of the two wavelengths, and the ratio of the intensity of reflected light of the two wavelengths is converted into the original transcutaneous bilirubin value through the built-in reflectance calculation module. The light source emitting end of the probe is designed to be tilted at an angle to the skin measurement surface.

3. The non-invasive neonatal jaundice monitoring method according to claim 2, characterized in that: The miniature optical imaging unit includes a multispectral ring light source and a high-resolution miniature camera. The multispectral ring light source is arranged around the probe measurement window and simultaneously emits a combination of visible light and near-infrared band illumination during light reflection measurement. The miniature camera captures images of the skin surface from a vertical perspective. The image resolution satisfies that the actual skin size corresponding to a single pixel is smaller than the minimum feature size of the vernix caseosa. The built-in image processing algorithm first performs distortion correction and illumination equalization preprocessing on the image, and then segments the vernix caseosa region through a convolutional neural network. The convolutional neural network uses the texture features and reflectivity of vernix caseosa as training labels.

4. The non-invasive neonatal jaundice monitoring method according to claim 3, characterized in that: When identifying the boundary between the vernix caseosa region and the non-vernix caseosa region, based on the binary segmentation map output by the convolutional neural network, morphological closing operations are used to fill the tiny pores in the vernix caseosa sheet, and then continuous closed boundaries are extracted by the edge detection algorithm. The vernix caseosa coverage area is obtained by counting the total number of pixels in the closed boundary. At the same time, isolated areas with an area smaller than the vernix caseosa fragment threshold are removed to eliminate measurement noise.

5. The non-invasive neonatal jaundice monitoring method according to claim 1, characterized in that: When matching the correction coefficient in the lookup table, the four nearest neighbor data cells are located using the vernix caseosa coverage value as the row index and the contact pressure value as the column index. If the vernix caseosa coverage value or the contact pressure value is between the two indices, the actual correction coefficient is calculated using bilinear interpolation. When applying the correction coefficient to the original transcutaneous bilirubin value, it is directly algebraically added to the original transcutaneous bilirubin value to generate an intermediate calibration value.

6. The non-invasive neonatal jaundice monitoring method according to claim 5, characterized in that: When calculating and outputting the corrected transcutaneous bilirubin value using linear superposition interpolation, a compensation factor related to the newborn's gestational age and measurement site is introduced for the intermediate calibration value. This compensation factor is pre-stored in the calibration model and its weight is automatically matched according to the input newborn's gestational age and measurement point location. The final corrected value is output as a weighted sum of the intermediate calibration value and the compensation factor to eliminate residual errors caused by individual physiological differences.

7. A non-invasive neonatal jaundice monitoring system, based on the non-invasive neonatal jaundice monitoring method according to any one of claims 1-6, characterized in that, include: The light reflection measurement unit (1) integrates a dual-wavelength light source module and a photoelectric sensor to execute the dual-wavelength red and blue light alternating emission mode. The light source emission end is arranged at an inclined angle to suppress vernix caseosa specular reflection. The reflectivity calculation module converts the ratio of the intensity of the received dual-wavelength reflected light into the original transdermal bilirubin value to provide initial input for calibration. The vernix caseosa analysis unit (2) consists of a miniature optical imaging unit and an image processing algorithm, including a multispectral ring light source, a high-resolution miniature camera and a convolutional neural network processor, used to capture skin surface images in real time, identify vernix caseosa region boundaries and calculate vernix caseosa coverage percentage, and output the quantified vernix caseosa coverage value to the calibration unit. The dynamic pressure sensing unit (3) is embedded in the piezoresistive sensor array of the probe contact surface to monitor the contact pressure value in real time. It integrates a pressure feedback mechanism. When the pressure deviates from the standard range, it triggers an audible and visual warning and pauses the calibration. After the pressure stabilizes, it activates the subsequent process. The intelligent calibration processing unit (4) has a built-in microprocessor and memory, which stores a three-dimensional lookup table and a compensation factor database. It is used to call the vernix caseosa coverage value, contact pressure value and original transcutaneous bilirubin value. It matches the correction coefficient through bilinear interpolation and superimposes the compensation factor to output the corrected transcutaneous bilirubin value that eliminates vernix caseosa interference and pressure fluctuation.

Images