Diabetes detection system based on multi-sensor pulse characteristics

Abstract

The invention discloses a diabetes detection system based on multi-sensor pulse characteristics, and relates to the field of medical signal processing, and the diabetes detection system comprises a pulse wave data acquisition module which uses a pressure sensor and a photoelectric sensor to collect wrist pressure pulse waves and fingertip photoelectric volume pulse waves of a user; the preprocessing module is used for obtaining usable single-cycle pulse wave data; the signal fusion module is used for converting the pressure pulse wave signal and the photoelectric volume pulse wave signal into two-dimensional image signals through continuous wavelet transformation so as to reserve time-frequency domain information, and processing two two-dimensional images by using non-subsampled shear wave transformation and a pulse coupling neural network; the feature extraction and calculation module is used for extracting pulse wave image features through a ResNet network and finally performing classification through a random forest; according to the detection system, the fusion effect is improved under the condition that the calculated amount is not increased, the effect of noninvasive detection of diabetes by using multiple sensors is achieved, and the detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical signal processing, and specifically, to a diabetes detection system based on multi-sensor pulse characteristics. Background Art

[0002] With the development of technology and the improvement of social level, the number of people suffering from diabetes in modern society is increasing. Diabetes has gradually been listed together with hypertension and cardiovascular diseases as the world's three major chronic diseases. On May 19, 2023, the "World Health Statistics Report 2023" pointed out that diabetes caused 2 million deaths globally in just one year. According to the data prediction of the authoritative medical journal "The Lancet" in 2023, in the next 30 years, the number of diabetes patients is expected to increase to 1.3 billion, and there will be growth in every country. Therefore, the timely detection of diabetes is particularly important today.

[0003] Currently, the main methods for detecting diabetes include fasting blood glucose test, oral glucose tolerance test, glycated hemoglobin test, and minimally invasive portable blood glucose meter, etc. The fasting blood glucose test requires the patient to fast and refrain from drinking water for more than eight hours, and then ingest 237 ml of glucose solution, which will cause discomfort to the patient; the oral glucose tolerance test requires the patient to orally take an aqueous solution containing 75 g of anhydrous glucose, and then collect blood at 0.5, 1.0, and 2.0 hours respectively to measure its blood glucose changes to observe the patient's ability to tolerate glucose. This will not only cause discomfort in the patient's stomach but also requires three blood collections; while both the glycated hemoglobin test and the minimally invasive portable blood glucose meter need to collect the patient's blood and are relatively expensive. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] Aiming at the problems that the existing diabetes detection methods are invasive and painful, and usually focus on a single signal for prediction, with limited information obtained and it is difficult to comprehensively evaluate the human body state. In order to overcome at least one deficiency in the prior art, this application provides a diabetes detection system based on multi-sensor pulse characteristics, which solves the problems raised in the above background art.

[0006] The present invention provides a diabetes detection system based on multi-sensor pulse characteristics, including:

[0007] S1. A pulse wave data acquisition module, which uses an array pressure sensor and a photoelectric sensor to collect multi-channel pulse wave data;

[0008] S2. A signal preprocessing module, which preprocesses the multi-channel pulse wave data, including signal filtering, baseline drift removal, and period segmentation, etc., to obtain usable single-cycle pressure pulse wave data and photoplethysmogram data;

[0009] S3. The signal fusion module converts the pulse wave data obtained in S2 into a two-dimensional image through continuous wavelet transform, uses non-subsampled shearlet transform (NSST) on the converted two-dimensional image to decompose the image into a low-frequency subband and high-frequency subbands, then fuses the low-frequency subband using the energy attribute weighted average method and fuses the high-frequency subband using a pulse coupled neural network (PCNN). Finally, an inverse shear transform is used to generate the fused image. The fused image using this method is superior in terms of detail retention, contrast, and structural integrity.

[0010] S4. The feature extraction and calculation module uses a ResNet network to extract features from the fused image and uses a random forest for feature classification to achieve the purpose of non-invasive and accurate detection of diabetes.

[0011] Furthermore, for the collection of the pulse wave signal of diabetic patients in S1, an integrated multi-channel collection device is adopted. This device precisely collects synchronous three-channel pressure pulse wave signals on the wrist of the subject through a three-channel array pressure sensor. At the same time, a high-performance photoelectric sensor is used to collect the pulse wave signal at the fingertip of the subject's right index finger.

[0012] The preprocessing module includes signal filtering, baseline drift removal, and period segmentation, etc. The signal filtering method is specifically manifested as sequentially passing each group of 3-channel pulse wave signals through an FIR low-pass filter with a cut-off frequency of fl to remove the high-frequency noise of the signal, respectively judging whether there is distortion or weak signal in each channel of the pulse wave through the amplitude and slope changes of the pulse wave, and detecting whether there is a sudden change in the signal by calculating the autocorrelation function to eliminate abnormal signals.

[0013] If the signal is periodic, the autocorrelation function (ACF) will show significant peaks at certain delays, and these peaks correspond to the period of the signal. On the contrary, obvious periodic peaks will not be shown.

[0014] Furthermore, the ascending branch part of the pulse wave is located in the cardiac systolic phase and has the most obvious fluctuation. Its first-order difference value is greater than zero throughout the pulse wave cycle. Therefore, the point with the largest first-order difference value of the ascending branch part of the pulse wave can be located. This point is easier to determine throughout the pulse cycle, and then search leftward for the waveform minimum point, that is, the starting point.

[0015] Furthermore, the method of cubic spline curve fitting is used to remove the baseline drift. The baseline is fitted based on the above starting point, the overall signal fluctuation is corrected by calculating the difference between the signal and the baseline, and finally, period segmentation processing is performed based on the above starting point.

[0016] The signal fusion module performs two-dimensional conversion on the preprocessed pressure pulse wave signal and photoplethysmogram signal using the continuous wavelet transform method.

[0017] Further, in the signal fusion part of S3, two-dimensional images of the pressure pulse wave and the photoplethysmogram are first input, and the images are denoised and aligned to ensure the spatial consistency between the images.

[0018] The non-subsampled shearlet transform (NSST) is respectively applied to the preprocessed two-dimensional images of the pressure pulse wave and the photoplethysmogram to decompose the images into low-pass subbands and high-pass subbands. The low-pass subbands contain the structural information of the images, and the high-pass subbands contain the detailed information of the images.

[0019] The formula is as follows:

[0020]

[0021] In the formula, j, k, and m are parameters in the shearlet transform, representing the decomposition level, direction, and spatial position respectively.

[0022] Further, S3 calculates the intrinsic property (IP) and energy attribute (EA) of each low-pass subband. IPA and IPB are the intrinsic properties of the low-pass subbands of the two-dimensional images of the pressure pulse wave and the photoplethysmogram, usually determined by the mean and median:

[0023] IPA = μA + μ′ A

[0024] IPB = μB + μ′ B

[0025] Among them, μ is the mean, μ′ is the median, and δ is a scaling factor.

[0026] The energy attribute is a method for measuring the pixel energy in the image subband and is used to reflect the structural characteristics of the image. For each pixel position (p, q), the energy attributes of the two images are calculated, and the formula is as follows:

[0027] EAA(p, q) = e -δ|LA(p,q)-IPA|

[0028] EAB(p, q) = e -δ|LB(p,q)-IPB|

[0029] LA(p, q) and LB(p, q) are the pixel values of the low-pass subbands of the two images at the position (p, q) respectively.

[0030] Fuse the low-frequency sub-bands using the Energy Attribute Weighted Average method, and the formula is as follows:

[0031]

[0032] Fuse the low-frequency sub-bands of the two-dimensional images of the pressure pulse wave and the photoplethysmogram into one low-frequency sub-band.

[0033] Further, in S3, the Pulse Coupled Neural Network (PCNN) is used to process the high-frequency sub-bands. The Pulse Coupled Neural Network (PCNN) is used to determine the optimal fusion rule for the two types of pulse information to retain important structural and texture information.

[0034] The formula is expressed as:

[0035]

[0036] In the formula, h[m] and g[m] are the coefficients of the low-pass filter and the high-pass filter respectively.

[0037] According to the output of the PCNN, select the high-frequency sub-band of the pressure pulse wave or the photoplethysmogram as the fused high-frequency sub-band.

[0038] Further, in S3, the Inverse Shearlet Transform is used to recombine the sub-bands that have undergone feature extraction and fusion processing to reconstruct the fused image, restore the integrity and consistency of the image, retain the structural information and functional information of the image, thereby providing more accurate details, while retaining the structural features of the input image, suppressing noise, and improving the overall quality of the fused image.

[0039] Further, input the generated fused image into the pre-trained ResNet network. Through the layer-by-layer processing of the convolutional layer, pooling layer, and residual module, extract the deep features of the image, and at the same time extract high-level semantic features, improve the ability to extract deep features, and help distinguish different image features.

[0040] Furthermore, the Random Forest (RF) algorithm is used to classify the extracted image features. Random Forest is an ensemble learning algorithm improved based on the decision tree algorithm. By constructing multiple decision trees and combining their prediction results, the final classification result is obtained. When constructing the Random Forest, first, a part of the samples is randomly selected from the training dataset as the training data for the current decision tree, and a part of the features is randomly selected from the feature set to construct the decision tree. Then, using the selected training data and features, the decision tree is constructed by splitting nodes until the stopping condition is met. Finally, the deep image features extracted from the ResNet network are input into the Random Forest model.

[0041] Each decision tree in the Random Forest makes a classification decision based on the input features and outputs its own classification result. Through the above system, diabetes detection based on multi-sensor pulse features is realized.

[0042] Compared with the prior art, the present invention provides a diabetes detection system based on multi-sensor pulse features, having the following beneficial effects:

[0043] The diabetes detection system based on multi-sensor pulse features provided by the present invention uses a pressure sensor and a photoelectric sensor to collect the pulse wave signals of users, without performing invasive operations on users, avoiding risks such as pain and infection that may be brought by traditional invasive detection methods, improving the acceptance of users and the convenience of detection, solving the problems of invasiveness and pain of existing diabetes detection methods, being applicable to users of different ages and different physical conditions, without the need for professional operators, having wide applicability and promotion value, and being suitable for home screening by patients.

[0044] The present invention uses continuous wavelet transform to convert the pulse wave signal into a two-dimensional image signal, which can better retain the time-frequency characteristics of the signal and provide a better data basis for subsequent image processing and feature extraction. At the same time, the non-subsampled shearlet transform (NSST) and the pulse-coupled neural network (PCNN) are used to process the two-dimensional image. Without significantly increasing the computational amount, the fusion effect is improved, important feature information in the image is effectively extracted, the signal recognizability is further enhanced, and the reliability of the detection result is improved.

[0045] By fusing the pressure pulse wave signal and the photoplethysmogram signal, the present invention can make full use of the advantages of the two sensor signals and make up for the deficiencies of a single sensor signal. The pressure pulse wave signal can reflect the pressure changes in the blood vessel wall, while the photoplethysmogram signal can reflect the changes in blood volume. The fusion of the two can more comprehensively reflect the physiological state of the human blood vessels and blood, thereby providing richer information for the detection of diabetes and effectively improving the accuracy of detection. This helps to improve the early diagnosis rate and treatment effect of diabetes and is of great significance for the prevention and control of diabetes. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 FIG. is a block diagram of a diabetes detection system based on multi-sensor pulse characteristics according to an embodiment of the present invention;

[0047] Figure 2 FIG. is a schematic diagram of the pulse image fusion process of the present invention;

[0048] Figure 3 FIG. is a structural diagram of the ResNet network of the present invention;

[0049] Figure 4 FIG. is a block diagram of the random forest structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] Embodiment

[0052] As Figure 1 shown, a diabetes detection system based on multi-sensor pulse characteristics has the following specific steps:

[0053] S1: Use an array pressure sensor and a photoelectric sensor to collect multi-channel pulse wave data.

[0054] In this embodiment, an integrated three-channel pressure sensor is used to collect the pressure pulse wave signals at the three positions of the ulnar, cun, and guan of the user's wrist, and a photoelectric sensor is used to collect the photoplethysmogram at the index finger of the user's right hand.

[0055] S2: Preprocess the multi-channel pulse wave data, including signal filtering, baseline drift removal, and period segmentation, etc., to obtain single-cycle three-channel pressure pulse wave data and single-channel photoplethysmogram data that can be used;

[0056] The specific steps are as follows:

[0057] The preprocessing module includes signal filtering, baseline drift removal, and period segmentation, etc. The signal filtering method is specifically manifested as follows: successively passing the 3-channel pulse wave signals of each group through an FIR low-pass filter with a cut-off frequency of fl to remove the high-frequency noise of the signals. The pulse wave judges whether there is distortion or weak signal in each channel of the pulse wave respectively through the amplitude and slope changes of the pulse wave, and detects whether there is a sudden change in the signal by calculating the autocorrelation function to eliminate abnormal signals.

[0058] If the signal is periodic, the autocorrelation function (ACF) will show significant peaks at certain delays, and these peaks correspond to the periods of the signal. On the contrary, obvious periodic peaks will not be shown.

[0059] Furthermore, the ascending branch part of the pulse wave is located in the cardiac systolic phase and has the most obvious fluctuations. Its first-order difference value is greater than zero throughout the pulse wave period. Therefore, the point with the largest first-order difference value of the ascending branch part of the pulse wave can be located. This point is easier to determine in the whole pulse period, and then look for the waveform minimum point to the left, that is, the starting point.

[0060] Furthermore, the method of cubic spline curve fitting is used to remove the baseline drift. The baseline is fitted based on the above starting point, and the overall signal fluctuation is corrected by calculating the difference between the signal and the baseline. Finally, the period segmentation process is carried out based on the above starting point.

[0061] S3. Signal fusion module, which fuses the pulse signals based on multiple sensors. The specific operation steps are as follows:

[0062] A1: Perform two-dimensional conversion on the pressure pulse wave signal and photoplethysmogram signal obtained by preprocessing using the continuous wavelet transform method.

[0063] A2: As Figure 2 shown, apply the nonsubsampled shearlet transform (NSST) to the two-dimensional images of the preprocessed pressure pulse wave and photoplethysmogram respectively, and decompose the images into low-pass subbands and high-pass subbands. The formula is as follows:

[0064]

[0065] In the formula, j, k, and m are the parameters in the shearlet transform, representing the decomposition level, direction, and spatial position respectively.

[0066] The low-pass subbands contain the structural information of the images, and the high-pass subbands contain the detail information.

[0067] A2: Further, due to the different information focuses of the pressure pulse wave and the photoplethysmogram, in this example, the energy attribute weighted average method (Energy Attribute Weighted Average) is used to fuse the low-frequency subbands, and the intrinsic property (IP) and energy attribute (EA) of each low-frequency subband are calculated.

[0068] IPA and IPB are the intrinsic properties of the two-dimensional image low-frequency subbands of the pressure pulse wave and the photoplethysmogram, usually determined by the mean and median. The formulas are as follows:

[0069] IPA = μ A + μ′ A

[0070] IPB = μ B + μ′ B

[0071] Where μ is the mean, μ′ is the median, and δ is a scaling factor used to control the range of the energy attribute. For each pixel position (p, q), the energy attributes of the two images are calculated. The formulas are as follows:

[0072] EAA(p, q) = e -δ|LA(p,q)-IPA|

[0073] EAB(p, q) = e -δ|LB(p,q)-IPB|

[0074] LA(p, q) and LB(p, q) are the pixel values of the two image low-frequency subbands at the position (p, q), respectively.

[0075] The energy attribute weighted average method is used to fuse the low-frequency subbands. The formula is as follows:

[0076]

[0077] By calculating the energy attribute (EA) of each low-frequency subband and performing weighted averaging on the subbands based on these attributes, fusion is achieved.

[0078] A3: The pulse-coupled neural network (PCNN) is used to process the high-frequency subbands to retain the detailed information.

[0079] The formula is expressed as:

[0080]

[0081] In the formula, h[m] and g[m] are the coefficients of the low-pass filter and the high-pass filter, respectively.

[0082] According to the output of the PCNN, select the high-frequency sub-band of the pressure pulse wave or the photoplethysmogram as the fused high-frequency sub-band.

[0083] A4: Use the Inverse Shearlet Transform to recombine the sub-bands that have undergone feature extraction and fusion processing, reconstruct the fused image, restore the integrity and consistency of the image, retain the structural and functional information of the image, and thus provide more accurate details.

[0084] S4. Feature extraction and classification recognition module: Extract the pulse wave image features from the fused image generated by S3 through the trained ResNet network, and classify them through a random forest, achieving the purpose of non-invasive detection of diabetes.

[0085] A1: First, use the ResNet network pre-trained on the prepared dataset as a feature extractor. The ResNet network is as Figure 3 shown. The network has learned rich image feature representations through training and directly uses the weight parameters learned in the convolutional layer, pooling layer, and residual module as a basis. Input the fused image obtained from S3 into the ResNet network, and extract the deep features of the image through the layer-by-layer processing of the convolutional layer, pooling layer, and residual module. The parameters of the ResNet network are shown in Table 1.

[0086] Table 1 ResNet network parameters

[0087]

[0088] A2: As Figure 4 shown, in this embodiment, the Random Forest (RF) algorithm is used to classify the extracted image features. When constructing a random forest, first randomly select a part of the samples from the training dataset as the training data for the current decision tree, and randomly select a part of the features from the feature set to construct the decision tree. Then, use the selected training data and features to construct the decision tree by splitting nodes until the stopping condition is met. Input the deep image features extracted from the ResNet network into the random forest model. Each decision tree in the random forest makes a classification decision based on the input features and outputs its own classification result. Through the above method, diabetes detection based on multi-sensor pulse features is achieved.

[0089] In this example, the MATLAB R2020a is used to perform 5-fold cross-validation on the dataset, randomly allocate 80% of the data as the training set, and the remaining 20% of the data as the test set. The model is retested five times, and the average value of the prediction is calculated.

[0090] The evaluation metrics for this example are as follows:

[0091] In this example, a confusion matrix is used to reflect the classification results of the classification model. TP represents that a diabetic sample is correctly diagnosed as diabetic by the classification model; FN represents that a diabetic sample is misdiagnosed as healthy by the classification model; FP represents that a healthy sample is misdiagnosed as diabetic by the classification model; TN represents that a healthy sample is correctly diagnosed as healthy by the classification model, as shown in Table 2.

[0092] Table 2 Confusion Matrix

[0093]

[0094] Based on the confusion matrix, this study further uses accuracy (A CC ), precision (P), recall (R), and F1-score as evaluation metrics to better evaluate the performance of the model.

[0095]

[0096] Recall represents the proportion of patients actually suffering from diabetes who are diagnosed. It can reflect the missed diagnosis rate. The higher the recall, the lower the missed diagnosis rate.

[0097]

[0098] The F-value is a comprehensive metric considering precision and recall.

[0099] When α = 1, it is the F1-score

[0100]

[0101] Aiming at the problems of the existing diabetes detection methods being invasive and painful, and usually focusing on a single signal for prediction, with limited information obtained and difficulty in comprehensively evaluating the human body state, in order to overcome at least one deficiency in the prior art, this application provides a diabetes detection system based on multi-sensor pulse characteristics, which solves the problems raised in the above background art.

[0102] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A diabetes detection system based on multi-sensor pulse characteristics, characterized in that, Including: A pulse wave data acquisition module that acquires multi-channel pulse wave data using an array pressure sensor and a photoelectric sensor; A signal preprocessing module that preprocesses the multi-channel pulse wave data, including signal filtering, baseline drift removal, and period segmentation, etc., to obtain usable single-period pressure pulse wave data and photoplethysmogram data; A signal fusion module that converts the obtained pulse wave data into a two-dimensional image through continuous wavelet transform, uses non-subsampled shearlet transform on the converted two-dimensional image to decompose the image into a low-frequency subband and a high-frequency subband, then uses the energy attribute weighted average method to fuse the low-frequency subband, and uses pulse coupled neural network to fuse the high-frequency subband, and finally uses inverse shear transform to generate the fused image; A feature extraction and calculation module that uses a ResNet network to extract features from the fused image and uses a random forest for feature classification to achieve the purpose of non-invasive and accurate detection of diabetes.

2. The diabetes detection system based on multi-sensor pulse characteristics according to claim 1, wherein: The pulse wave data acquisition module uses an integrated multi-channel acquisition device for collecting the pulse wave signal of the user; this device uses a three-channel array pressure sensor to accurately collect the synchronous three-channel pressure pulse wave signal of the subject's wrist; at the same time, a high-performance photoelectric sensor is used to collect the pulse wave signal of the subject.

3. The diabetes detection system based on multi-sensor pulse characteristics according to claim 1, characterized in that: The signal preprocessing module sequentially removes the high-frequency noise of each group of 3-channel pulse wave signals through an FIR low-pass filter with a cut-off frequency of fl, respectively judges whether there is distortion or weak signal in each channel of the pulse wave through the amplitude and slope changes of the pulse wave, and detects whether the signal is aperiodic by calculating the autocorrelation function, and eliminates abnormal signals.

4. The diabetes detection system based on multi-sensor pulse characteristics according to claim 1, characterized in that: The signal fusion module converts the preprocessed pressure pulse wave signal and photoplethysmogram signal into a two-dimensional image using the continuous wavelet transform method to retain the time-domain and frequency-domain information of the pressure pulse wave signal and photoplethysmogram signal, and facilitate multi-sensor signal fusion.

5. The diabetes detection system based on multi-sensor pulse characteristics according to claim 1, wherein: The signal fusion module uses non-subsampled shearlet transform to decompose the image into a low-frequency subband and a high-frequency subband. The low-frequency subband contains the structural information of the image, and the energy attribute weighted average method is used to fuse the low-frequency subband; the high-frequency subband contains the detail information, and the pulse coupled neural network is used to fuse the high-frequency subband, and then the fused image is generated through inverse shear transform. The fused image using this method is superior in terms of detail retention, contrast, and structural integrity.

6. The diabetes detection system based on multi-sensor pulse characteristics according to claim 1, wherein: The feature extraction and calculation module performs feature extraction and calculation on the generated fused image for classification. Specifically, it extracts the pulse wave image features from the fused image generated by the signal fusion module through a trained ResNet network and classifies them through a random forest, achieving the purpose of non-invasive detection of diabetes.

Images