Health monitoring method based on combination of body fluid SERS (Surface Enhanced Raman Scattering) spectrum and B ultrasonic image

Abstract

The invention discloses a health monitoring method based on combination of body fluid SERS spectrum and B ultrasonic image, comprising the following steps: S1, collecting data of a monitored person, the data comprising physiological indexes, urine and carotid artery ultrasonic image, processing the urine to obtain SERS signals, processing the SERS signals to obtain first feature data, and processing the first feature data to obtain second feature data; processing the carotid artery ultrasonic image through an Alexnet model framework to obtain second feature data; s2, taking the first feature data and the second feature data as independent data, performing model prediction to obtain a first probability based on the first feature data and a second probability based on the second feature data, then taking the obtained first probability and second probability as new independent parameters, combining the new independent parameters with the physiological indexes, and performing model prediction again to obtain a model prediction result; when the output result is 1, the risk of suffering from the cerebral arterial thrombosis is indicated. And the SERS spectrum of the body fluid and the B ultrasonic image are combined for judgment, so that the accuracy of a monitoring result can be effectively improved, and a monitor can conveniently and further see a doctor for treatment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of health monitoring, and particularly relates to a health monitoring method based on combination of body fluid SERS spectrum and B-ultrasound image. BACKGROUND

[0002] With the intensification of global aging trend, the demand for family health monitoring and early diagnosis of diseases is increasing, and the technology of fusing multi-modal biological information (such as image, genome, body fluid marker, etc.) becomes the core solution to crack the medical pain points of aging. However, traditional diagnostic methods have significant limitations and are difficult to meet the current needs. Taking ischemic stroke (IS) with high disability rate and high mortality rate as an example: traditional imaging technology (CT / MRI / DSA) has defects such as insufficient sensitivity in acute stage (such as super-early false negative), risk of invasive operation and lack of real-time monitoring, and the limited action of IS patients and the lack of basic medical resources further hinder timely diagnosis. SUMMARY

[0003] The application proposes a health monitoring method based on combination of body fluid SERS spectrum and B-ultrasound image, which judges the risk of suffering from IS by combining SERS spectrum and B-ultrasound image multi-modal data, so as to facilitate the monitor to seek medical treatment and transfer to hospital for treatment in time.

[0004] To this end, the technical scheme adopted by the application is as follows: a health monitoring method based on combination of body fluid SERS spectrum and B-ultrasound image, comprising the following steps:

[0005] S1: data acquisition; collecting data of the monitor, the data including physiological indicators, urine and carotid artery ultrasound image, wherein the urine is processed to obtain SERS signal, and the SERS signal is subjected to denoising and abnormality processing to obtain first feature data, and the carotid artery ultrasound image is processed by an Alexnet model framework to output second feature data of a certain size;

[0006] S2: data processing, taking the first feature data and the second feature data as independent data, performing model prediction to obtain a first probability based on the first feature data and a second probability based on the second feature data, then taking the obtained first probability and second probability as new independent parameters, combining them with the physiological indicators, and performing model prediction again, when the output result is 1, it indicates that there is a risk of suffering from ischemic stroke.

[0007] As preferred in the above-mentioned solution, in S1, the processing of urine includes the following steps: 5-10 mL of midstream morning urine is taken and placed in a sterile centrifuge tube, and transported from the hospital to the laboratory at a temperature of 4℃, and then divided into two parts, one of which is stored at -80℃ for long-term preservation, and the other is first treated by centrifugation, then the supernatant is added with PBS buffer and placed at -30℃, and then the signal acquisition is completed within three days;

[0008] The process of signal acquisition includes taking a certain solution and dropping it on the SERS chip, and drying it on a 40℃ hot plate, and collecting it by a Raman spectrometer, and when collecting, a laser with a wavelength of 532 nm and 633 nm is respectively focused on the sample point by a 50x objective, and the collection time is 1s, and the SERS spectrum is obtained 532 and SERS 633 spectrum.

[0009] Further preferably, the SERS chip is prepared by the following steps: the surface of the organic nylon filter paper blank substrate is activated and the hydrophilicity is improved by using the plasmonic surface modification method, then the filter paper is immersed in an ethanol solution of 3-aminopropyl triethoxysilane for 2h to functionalize the surface, wherein the volume ratio of 3-aminopropyl triethoxysilane to ethanol is 1:20, and the filter paper is washed thoroughly with ethanol, then 110nm silver nanoparticle colloidal solution is loaded onto the filter paper by pressure filtration, the substrate is washed with deionized water and dried, and finally the paper substrate is cut into 0.5cm x 0.5cm.

[0010] Further preferably, the carotid ultrasound image is collected as follows: first, in the transverse section, the right side starts from the beginning of the innominate artery, and the left side starts from the branch of the common carotid artery from the aortic arch, and the whole process of the bilateral common carotid artery, the bifurcation of the internal and external carotid arteries, the initial segment of the internal carotid artery and the range of 4-6cm beyond the bifurcation, the three-layer structure of the blood vessel wall and the presence or absence of atherosclerotic plaques in the blood vessel lumen of the main stem of the external carotid artery and its branches are observed continuously, and then the intima thickness of the distal segment of the common carotid artery is measured in the longitudinal section.

[0011] Further preferably, the physiological indicators include hypertension, atrial fibrillation, diabetes, and smoking history data.

[0012] Further preferably, in S2, the model used by the first feature data is support vector machine, and the data combination used is SERS spectrum 532 , SERS 633 spectrum and physiological indicators.

[0013] Further preferably, in S2, the model used by the second feature data is support vector machine, and the data combination used is ultrasound image and physiological indicators.

[0014] It is further preferred that the model adopted by the first probability, the second probability and the physiological index in S2 is a Gaussian naive Bayes model.

[0015] Advantages of the present application: the combination of the body fluid SERS spectrum and the B-ultrasound image can effectively improve the accuracy of the monitoring result, facilitate the further medical treatment of the monitor, and the SERS spectrum and the B-ultrasound image both belong to non-invasive methods, which are simple to operate. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the present application.

[0017] Figure 2 The physiological index data of the recruited population.

[0018] Figure 3 The SERS spectrum of the recruited population.

[0019] Figure 4 The carotid ultrasound image of the recruited population.

[0020] Figure 5 The data processing method in the present application.

[0021] Figure 6 The verification of the SVM model.

[0022] Figure 7 The verification of the GNB model. DETAILED DESCRIPTION

[0023] The present application will be further described below by way of examples and in conjunction with the drawings:

[0024] As shown in the figure, a health monitoring method based on the combination of body fluid SERS spectrum and B-ultrasound image, comprising the following steps: Figures 1-7

[0025] First step: data acquisition. The data of the monitor is collected, including physiological index, urine and carotid ultrasound image, wherein the urine is treated to obtain SERS signal, and the SERS signal is subjected to noise reduction and abnormality elimination to obtain first feature data, and the carotid ultrasound image is subjected to Alexnet model framework processing to output second feature data of a certain size. The SERS signal collected is subjected to noise reduction processing by Savitzky-Golay filtering, baseline correction by airPLS, normalization by Min-Max normalization, and elimination of abnormal data by interquartile range (IQR) method. Savitzky-Golay filtering, airPLS, Min-Max normalization, interquartile range (IQR) method and Alexnet model framework are all prior art, and the detailed process will not be described here. ​

[0026] For IS monitoring, the physiological indicators are hypertension, atrial fibrillation, diabetes, and smoking history data.

[0027] The urine processing includes the following steps: 5-10 mL of the middle section of the first morning urine is taken into a sterile centrifuge tube every day, and is transported from the hospital to the laboratory at a temperature of 4℃, and then is divided into two parts, one of which is long-term stored at -80℃, and the other is first treated by centrifugation, then the supernatant is added with PBS buffer solution and placed in -30℃, and then signal acquisition is completed within three days. The ratio between the supernatant and the PBS buffer solution is 1:4, and the PBS buffer solution is a phosphate buffered saline solution. When the signal is collected, the supernatant added with the PBS buffer solution is thawed at 4℃.

[0028] The signal acquisition process includes dropping a certain solution on the SERS chip and drying it on a 40℃ hot plate, and collecting it by a Raman spectrometer. When collecting, a laser with a wavelength of 532 nm and 633 nm is respectively focused on the sample point by 50x, the collection time is 1s, and the SERS spectrum is obtained 532 and SERS 633 spectrum. To ensure the consistency of the obtained spectrum, the signal is collected in a 10x10 array with an interval of 20μm. To ensure that the entire surface of the SERS chip is covered, 15μL of solution is dropped on the SERS chip each time.

[0029] The SERS chip is prepared by the following steps: the surface of the organic nylon filter paper blank substrate is activated and the hydrophilicity is improved by using the plasmonic surface modification method, then the filter paper is immersed in an ethanol solution of 3-aminopropyl triethoxysilane for 2h to functionalize the surface, the volume ratio of 3-aminopropyl triethoxysilane to ethanol is 1:20, and the substrate is washed thoroughly with ethanol, then 110nm silver nanoparticle colloidal solution is loaded onto the filter paper by pressure filtration, the substrate is washed with deionized water and dried, and finally the paper substrate is cut into 0.5cmx0.5cm.

[0030] The silver nanoparticles are prepared by the following method: first, 30 mL of ethylene glycol (EG) solution is added to a 100 mL round-bottom flask, and then the round-bottom flask is transferred to an oil bath at 165°C, stirred and heated for 20 minutes. Then, 0.08 mL of NaHB4 solution (concentration 0.015 g / mL), 1.5 mL of PVP solution (concentration 0.02 g / mL) and 0.4 mL of AgNO3 solution (concentration 0.048 g / mL) are sequentially added to the flask under stirring, and heated for 20 minutes. The solution color changes from colorless to brick red, and finally to orange yellow, forming round silver nanoparticles with a diameter of about 110 nm. Finally, the obtained solution is washed with ethanol three times (10000 rpm, 10 minutes), the solid components are dispersed in ethanol, and stored at room temperature for later use.

[0031] In urine detection, the silver nanoparticles are loaded on paper, and the patient's urine is directly dropped on the paper base for detection, so that the detection material is convenient to move and store, and the detection method is simple.

[0032] The carotid ultrasound image is collected as follows: first, in the transverse section, the right side starts from the brachiocephalic artery, and the left side starts from the carotid artery branch of the aortic arch, continuously observing the whole bilateral common carotid artery, the bifurcation of the internal and external carotid arteries, the initial segment of the internal carotid artery and the bifurcation beyond 4-6 cm, the three-layer structure of the blood vessel wall and the blood vessel lumen of the main stem of the external carotid artery and its branch blood vessels, and whether there is an atherosclerotic plaque, and then the intima-media thickness (IMT) of the distal common carotid artery (1.0-1.5 cm range below the bifurcation level) is measured in the longitudinal section.

[0033] Second step: data processing. The first feature data and the second feature data are used as independent data for model prediction to obtain a first probability (P1) based on the first feature data and a second probability (P2) based on the second feature data, and then the obtained first probability and second probability are used as new independent parameters, combined with physiological indicators for model prediction again, and when the output result is 1, it indicates that there is a risk of suffering from ischemic stroke, such as Figure 5 A shown in the middle.

[0034] Specifically, the model used for the first feature data is support vector machine (SVM), the data combination used is SERS spectrum 532 , SERS 633 spectrum and physiological indicators, the model used for the second feature data is support vector machine (SVM), the data combination used is ultrasound image and physiological indicators, and the model used for the first probability, the second probability and the physiological indicators is Gaussian naive Bayes model (GNB).

[0035] To verify the reliability of the method for monitoring, the following operations are taken for verification:

[0036] Step 1: Data collection and preliminary processing.

[0037] A cohort of 101 participants was recruited, including 55 healthy controls (HC) and 46 IS patients.

[0038] 1) Physiological parameters

[0039] The distribution of physiological parameters showed significant differences between HC and IS patients in hypertension, atrial fibrillation, diabetes, and smoking history (P<0.05), and IS patients were more likely to have these conditions, as shown in Figure 2 .

[0040] 2) Collection of SERS signals in urine

[0041] For urine monitoring, the biomolecules and chemicals in urine change over time, which affects the resulting Raman signals. Therefore, signal collection within 3-6 h after urine collection can minimize signal interference in label-free SERS detection. When collecting signals, different excitation wavelengths are expected to have different spectral profiles due to the combined effects of excitation wavelength-molecular absorption matching (SERRS dominant) and molecular-substrate chemical bonding patterns (CT dominant). Since the signal molecules in urine are not selected, some signals may not effectively reflect the characteristics of the sample, so 100 signals from each sample were scanned to capture as many characteristic signals of biomolecules in urine as possible, as shown in Figure 3 . A total of 10100 urine SERS spectra were collected (5500 HC and 4600 IS). Among them Figure 3 a is the spectrum at a wavelength of 633 nm, Figure 3 a is the spectrum at a wavelength of 532 nm, Figure 3 c is the combined spectrum of the two.

[0042] 3) Carotid ultrasound image collection

[0043] Carotid ultrasound examination was performed by Mindray new generation of high-end portable color Doppler ultrasound M11 with high-frequency probe (L12-4s, 12 MHz) bedside, the focus position was fixed at the level of carotid artery, the image depth was uniformly adjusted to 3.5 cm. All study participants received B-mode ultrasound examination, and C-IMT was measured. Subsequently, the maximum longitudinal section was selected to collect carotid plaque ultrasound images. Since there may be plaques in the carotid arteries of some HC, in order to minimize the influence of plaques on the diagnostic results, 2-8 imaging pictures were collected for each study participant. Finally, all ultrasound images were processed by the Alexnet model framework, and the output feature data size was (4096, ). See Figure 4 . A total of 481 ultrasound images (129 HC and 352 IS) were collected.

[0044] Step 2: Data processing.

[0045] 1. Multiple combinations of three-dimensional data were tried, including: SERS 633 spectrum, SERS spectrum 532 , first combined spectrum (SERS spectrum 532 and SERS spectrum 633 ), second combined data (physiological indicators, SERS spectrum 532 and SERS spectrum 633 ), ultrasound images, and third combined data (ultrasound images + physiological indicators). Various ML models were used to diagnose and evaluate IS patients, including: LGB Classifier (LGB), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Random Forest (RF), Gaussian Naive Bayes (GNB), eXtreme Gradient Boosting (XGB), and Alex Net. The best data and the best model were selected by five-fold cross-validation (the ratio of training set to test set was 8:2), as shown in B of Figure 5 . For urine SERS spectrum, the best data was the second combined data (physiological indicators, SERS spectrum 532 and SERS spectrum 633 ), and the best model was SVM, with an average accuracy of 0.84 and an area under the curve (AUC) of 0.91. For carotid ultrasound images, the best data was the third combined data (ultrasound images + physiological indicators), and the best model was SVM, with an average accuracy of 0.82 and an AUC of 0.84.

[0046] To evaluate the stability of the model, the dataset was randomly resampled (train set: test set = 1:1) and the best classification model was trained 3 times, respectively, to verify its diagnostic performance for IS. In the HC sample, the values of most signals were close to 0, while in the IS sample, the opposite was true. The average score of the IS patient sample was 8.1 times higher than that of the HC sample (SERS spectra) and 3.2 times higher (ultrasound images), respectively, with significant inter-group differences between the HC sample, see Figure 6 A. At the same time, the receiver operating characteristic (ROC) curve was calculated to evaluate the diagnostic performance. For carotid ultrasound images, the average accuracy and AUC reached 0.81 and 0.86, respectively, see Figure 6 B. For urine SERS spectra, the average accuracy and AUC reached 0.79 and 0.88, respectively, see Figure 6 C.

[0047] 2, the first probability based on urine SERS spectra (P1), the second probability based on ultrasound images (P2) and the screened 4 physiological indicators were fused to diagnose and evaluate IS patients. The GNB model obtained the best recognition result, with an accuracy of 0.92, an AUC of 0.95, a sensitivity of 0.86 and a specificity of 0.97, indicating that the comprehensive performance of this dataset combined with the GNB model is excellent, suitable for high-precision classification scenarios, see Figure 7 A. The F1 score of the precision-recall curve (PRC) and the area under the PRC (AUPRC) value reached 0.91 and 0.95, respectively, indicating that the model can well predict positive samples, see Figure 7 B. To verify the rationality of this research scheme, another multi-dimensional data fusion method (i.e. the fourth combined data, which includes ultrasound images, physiological indicators and SERS spectra) was used to diagnose and evaluate IS patients, and the results are shown in Table 1, with the accuracy and AUC of the best recognition result being 0.87 and 0.9, etc., which are lower than this method.

[0048] Table 1 Comparison of results of different data processing solutions

[0049] Accuracy Precision Sensitivity (Recall) Specificity F1 Score AUC PI + SERS 0.85 0.88 0.75 0.92 0.81 0.92 PI + CBI 0.85 0.87 0.90 0.73 0.89 0.88 PI + SERS + CBI 0.87 0.89 0.91 0.78 0.90 0.90 The method 0.92 0.96 0.86 0.97 0.91 0.95

Claims

1. A health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging, characterized in that, Includes the following steps: S1: Data Acquisition; Collects data from the monitor, including physiological indicators, urine and carotid ultrasound images. The urine is processed to obtain SERS signals, and then the SERS signals are denoised and abnormalities are removed to obtain the first feature data. The carotid ultrasound images are processed by the Alexnet model framework to output second feature data of a certain size. S2: Data processing; The first feature data and the second feature data are treated as independent data, and model prediction is performed to obtain the first probability based on the first feature data and the second probability based on the second feature data. Then, the obtained first probability and second probability are used as new independent parameters and combined with physiological indicators to perform model prediction again. When the output result is 1, it indicates that there is a risk of ischemic stroke.

2. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that, In S1, the urine processing includes the following steps: Take 5-10 mL of the midstream of each morning urine and put it into a sterile centrifuge tube. Transfer it from the hospital to the laboratory at 4°C. Then divide it into two portions. One portion is stored at -80°C for a long time. The other portion is centrifuged first, and the supernatant is added to PBS buffer and placed at -30°C. Then, signal acquisition is completed within three days. The signal acquisition process involves dropping a certain amount of solution onto the SERS chip, drying it on a 40°C heating plate, and then acquiring the signal using a Raman spectrometer. During acquisition, lasers with wavelengths of 532 nm and 633 nm are focused onto the sample points using a 50× objective lens, and the acquisition time is 1 second, yielding the SERS spectrum. 532 and SERS 633 spectrum.

3. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 2, characterized in that, The SERS chip was prepared by the following steps: the surface of the organic nylon filter paper blank substrate was activated and its hydrophilicity was improved by plasma surface modification. Then, the filter paper was immersed in an ethanol solution of 3-aminopropyltriethoxysilane for 2 hours to functionalize the surface, wherein the volume ratio of 3-aminopropyltriethoxysilane to ethanol was 1:

20. The paper was thoroughly rinsed with ethanol. Then, a colloidal solution of 110 nm silver nanoparticles was loaded onto the filter paper by pressure filtration. The substrate was rinsed with deionized water and dried. Finally, the paper substrate was cut into 0.5 cm × 0.5 cm pieces.

4. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that, The acquisition of carotid artery ultrasound images is as follows: First, in the transverse section, starting from the brachiocephalic artery on the right and from the branch of the common carotid artery to the aortic arch on the left, the entire length of the bilateral common carotid artery, the bifurcation of the internal carotid artery and external carotid artery, the initial segment of the internal carotid artery and the area 4-6 cm distal to the bifurcation, the three-layer structure of the main trunk and branches of the external carotid artery, and the presence of atherosclerotic plaques in the lumen of the vessel are observed. Then, in the longitudinal section, the carotid intima-media thickness of the distal segment of the common carotid artery is measured.

5. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that: The physiological indicators include data on hypertension, atrial fibrillation, diabetes, and smoking history.

6. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that: In S2, the first feature data uses a support vector machine model.

7. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that: In S2, the second feature data uses a support vector machine model.

8. The health monitoring method based on the combined use of body fluid SERS spectroscopy and B-ultrasound imaging as described in claim 1, characterized in that: In S2, the model used for the first probability, the second probability, and the physiological indicators is the Gaussian Naive Bayes model.

Images