Prediction Of Glucose Concentration Research Articles

A novel neural network architecture utilizing parametric-logarithmic-modulus-based activation function: Theory, algorithm, and applications

This paper introduces a novel parametric-logarithmic-modulus-based activation function (PLM-AF) designed to significantly enhance the nonlinear expression capabilities of high-dimensional spectroscopy data. A one-dimensional CNN-LSTM (1D-CNN-BiLSTM) model is subsequently developed to capture long-term dependencies within glucose Raman spectroscopy. To the best of our knowledge, this is the first work to simultaneously optimize the predictive performance of the model from the perspectives of both network architecture and activation functions. The effectiveness of the model is comprehensively evaluated against state-of-the-art methods using a public Raman spectroscopy dataset. Compared to the sub-optimal glucose prediction models, the proposed model improves the training root mean square error (RMSE) by 41.89%. The improved prediction accuracy demonstrates that the proposed regression model with the novel PLM-AF can significantly facilitate non-invasive glucose concentration prediction, thereby advancing the auxiliary diagnosis and healthcare industry.

Auto recalibration based on dual-mode sensing for robust optical continuous glucose monitoring

Boronic acid based optical continuous glucose monitoring systems (CGMs) hold promise for long-term glucose monitoring. However, these sensors suffer drifts induced inaccuracy and external calibration. One of the reasons is the denature of boronic acid structures and optical reporters. To solve this problem, a dual-mode glucose sensing method with auto-recalibration capabilities is developed using glucose-selective diboronic acid molecule (DBA) and optical reporter alizarin red S (ARS). DBA binds ARS at a 1:2 ratio and produces distinct changes in both the absorbance (Abs) and the fluorescence (FL) spectra of ARS. Through competitive binding with DBA, glucose induces a compensatory shift in both Abs and FL signals. The two signals were demonstrated to be strongly correlated and were combined to build Abs-FL-glucose 3D calibration curves for glucose determination. The mapping relationship of two signals drift closely with both changes of DBA and ARS concentration. By ascertain the right mapping relationship, the two strongly correlated signals can alert undesirable degradation of sensing components, identify the correct calibration curve, and accurately predict glucose concentration without external calibration. The introduction and adoption of the auto-recalibrating dual-mode sensing strategy could constitute a useful development in the pursuit of next generation long-term calibration-free CGM and wearable sensors technologies with superior precision.

Clinical decision support system based on artificial intelligence for adjusting insulin pump parameters in children with type 1 diabetes mellitus

BACKGROUND: Widely available diabetes devices (continuous glucose monitoring, insulin pump etc.) generate large amount of data and development of an advanced clinical decision support system (CDSS), able to automatically evaluate and optimize insulin therapy, is relevant.AIM: Development of a mathematical model and an CDSS based on it to optimize insulin therapy in children with type 1 diabetes (T1D) and assessment of the agreement between the recommendations of the CDSS and the physician on insulin pump (IP) parameters: basal profile (BP), carbohydrate ratio (CR), correction factor (СF).MATERIALS AND METHODS: Data from 504 children with T1DM were analyzed over the period of 7875 days. The data included glucose, insulin, food, sex, age, height, weight, diabetes duration and HbA1c. We constructed recurrent neural network (RNN) to predict glucose concentration for 30-120 minutes, an algorithm for optimizing IP settings using prediction results. Next, a software product was developed — a CDSS. To assess the agreement of the recommendations of the CDSS and physicians, retrospective data from 40 remote telemedicine consultations of 40 patients with T1D (median age 11.6 years [7; 15]) were used and 960 points of possible adjustments were analyzed. Three degrees of agreement have been introduced: complete agreement, partial agreement, and complete disagreement. The magnitude of the adjustments was also analyzed.RESULTS: The accuracy of glycemic predictions was better or comparable with other similar models. The assessment of agreement for BP, CR and CF, according to the Kappa index, showed slight and weak agreement. The frequency of complete agreement between recommendations for adjusting the ongoing IP therapy between the CDSS and physicians is 37.5–53.8%, and complete inconsistency is 4.5–17.4%. From a clinical point of view, consistency in the frequency of occurrence of the indicator is more important. There were no differences in median IP settings between the CDSS and physicians.CONCLUSION: The CDSS has an acceptable accuracy of glycemic predictions. The CDSS and physicians provide comparable recommendations regarding CSII parameters.

First-trimester HbA1c in relation to plasma glucose concentrations in an oral glucose tolerance test at 12 to 16 weeks’ gestation—a population-based study

BackgroundEarly-onset GDM often requires pharmacological treatment and is associated with adverse perinatal outcomes, but data is insufficient regarding the best methods to identify high-risk women requiring early GDM screening. The aim of this study was to analyze the diagnostic accuracy of HbA1c in the prediction of (1) plasma glucose concentrations > 90th percentile in an oral glucose tolerance test (OGTT) at 12–16 weeks’ gestation; and (2) pharmacologically treated early- or late-onset GDM.MethodsHbA1c was measured at 8—14 weeks’ gestation in a population-based cohort of 1394 Finnish women recruited for the Early Diagnosis of Diabetes in Pregnancy (EDDIE) study between 3/2013 and 12/2016. Information on maternal risk factors were collected at recruitment. Subsequently, a 2-hour 75 g OGTT was performed at 12—16 weeks’ gestation (OGTT1), and if normal, repeated at 24–28 weeks’ gestation (OGTT2). Early- and late-onset GDM were diagnosed using the same nationally endorsed cut-offs for fasting, 1 h- and 2 h-plasma glucose: ≥5.3, ≥ 10.0mmol/l, and/or ≥ 8.6mmol/l, respectively. In total, 52/1394 (3.7%) women required metformin or insulin treatment for GDM, including 39 women with early-onset GDM diagnosed at OGTT1 and 13 women with late-onset GDM diagnosed at OGTT2.ResultsMaternal early-pregnancy HbA1c ≥ 35mmol/mol (≥ 5.4%) was the best cut-off to predict fasting or post-load plasma glucose > 90th percentile in OGTT1, but its diagnostic accuracy was low [AUC (95% CI) 0.65 (0.62 to 0.69), sensitivity 0.55 (0.49 to 0.60) and specificity 0.67 (0.64 to 0.70)] both alone and in combination with other maternal risk factors. However, HbA1c ≥ 35mmol/mol correlated positively with plasma glucose concentrations at all time points of OGTT1 and predicted pharmacologically treated GDM diagnosed at OGTT1 or OGTT2; AUC (95% CI) 0.75 (0.68 to 0.81), sensitivity 0.75 (0.61 to 0.86), specificity 0.64 (0.61 to 0.66).ConclusionsIn our population-based cohort, early-pregnancy HbA1c ≥ 35mmol/mol was positively associated with fasting and post-load plasma glucose concentrations in an OGTT at 12—16 weeks’ gestation and predicted pharmacologically-treated early- and late-onset GDM, suggesting potential utility in first-trimester identification of women at high risk of severe GDM subtypes.

Validation of the cell culture monitoring using a Raman spectroscopy calibration model developed with artificially mixed samples and investigation of model learning methods using initial batch data.

The development of calibration models using Raman spectra data has long been challenged owing to the substantial time and cost required for robust data acquisition. To reduce the number of experiments involving actual incubation, a calibration model development method was investigated by measuring artificially mixed samples. In this method, calibration datasets were prepared using spectra from artificially mixed samples with adjusted concentrations based on design of experiments. The precision of these calibration models was validated using the actual cell culture sample. The results showed that when the culture conditions were unchanged, the root mean square error of prediction (RMSEP) of glucose, lactate, and antibody concentrations was 0.34, 0.33, and 0.25g/L, respectively. Even when variables such as cell line or culture mediawere changed, the RMSEPs of glucose, lactate, and antibody concentrations remained within acceptable limits, demonstrating the robustness of the calibration models with artificially mixed samples. To further improve accuracy, a model training method for small datasets was also investigated. The spectral pretreatment conditions were optimized using error heat maps based on the first batch of each cell culture condition and applied these settings to the second and third batches. The RMSEPs improved for glucose, lactate, and antibody concentration, with values of 0.44, 0.19, and 0.18g/L under constant culture conditions, 0.37, 0.12, and 0.12g/L for different cell lines, and 0.26, 0.40, and 0.12g/L when the culture media was changed. These results indicated the efficacy of calibration modeling with artificially mixed samples for actual incubations under various conditions.

Efficient Biofuel Generation of Plants via Enzymes

The purpose of this research is to observe the Efficiency of Biofuel Generation of Plants Using Enzymes. The question being asked is at which concentration of glucose will yeast fermentation rates be the most efficient in order to produce Ethanol through fermentation? It was predicted glucose concentrations and ethanol production rates would increase simultaneously. Ethanol is a renewable transportation fuel made from plants. Methods for researching this topic includes studying the environmental impacts of Ethanol, how it is produced, the process of enzyme digestion, the differences between cellulose and cellulase and the process of yeast alcohol fermentation. The methodology required for this experiment includes creating eight samples: Cellulase + Yeast, Cellulose + Yeast, Cellulase + Cellulose + Yeast, Cellulase + Cellulose + No yeast, and all four concentrations of glucose in separate samples (0.5%, 1%, 3% and 5%). The apparatus used to collect the data is made up of a syringe attached to a pipette, each sample started at a given volume of mL then as the samples began fermenting and releasing CO2 data was recorded for 15 minutes in 1 minute intervals.The volume of carbon dioxide released corresponds to the amount of sugar (glucose) fermented. The results of this experiment show a correlation between the increased rate of ethanol production as glucose concentrations increase. The interaction between cellulose and cellulase demonstrate the process of producing glucose through enzyme digestion.

Prediction of Blood Risk Score in Diabetes Using Deep Neural Networks.

Improving the prediction of blood glucose concentration may improve the quality of life of people living with type 1 diabetes by enabling them to better manage their care. Given the anticipated benefits of such a prediction, numerous methods have been proposed. Rather than attempting to predict glucose concentration, a deep learning framework for prediction is proposed in which prediction is performed using a scale for hypo- and hyper-glycemia risk. Using the blood glucose risk score formula proposed by Kovatchev et al., models with different architectures were trained, including, a recurrent neural network (RNN), a gated recurrent unit (GRU), a long short-term memory (LSTM) network, and an encoder-like convolutional neural network (CNN). The models were trained using the OpenAPS Data Commons data set, comprising 139 individuals, each with tens of thousands of continuous glucose monitor (CGM) data points. The training set was composed of 7% of the data set, while the remaining was used for testing. Performance comparisons between the different architectures are presented and discussed. To evaluate these predictions, performance results are compared with the last measurement (LM) prediction, through a sample-and-hold approach continuing the last known measurement forward. The results obtained are competitive when compared to other deep learning methods. A root mean squared error (RMSE) of 16 mg/dL, 24 mg/dL, and 37 mg/dL were obtained for CNN prediction horizons of 15, 30, and 60 min, respectively. However, no significant improvements were found for the deep learning models compared to LM prediction. Performance was found to be highly dependent on architecture and the prediction horizon. Lastly, a metric to assess model performance by weighing each prediction point error with the corresponding blood glucose risk score is proposed. Two main conclusions are drawn. Firstly, going forward, there is a need to benchmark model performance using LM prediction to enable the comparison between results obtained from different data sets. Secondly, model-agnostic data-driven deep learning models may only be meaningful when combined with mechanistic physiological models; here, it is argued that neural ordinary differential equations may combine the best of both approaches. These findings are based on the OpenAPS Data Commons data set and are to be validated in other independent data sets.

Large maternal waist circumference in relation to height is associated with high glucose concentrations in an early-pregnancy oral glucose tolerance test: A population-based study.

To explore the role of maternal anthropometric characteristics in early-pregnancy glycemia, we analyzed the associations and interactions of maternal early-pregnancy waist circumference (WC), height and pre-pregnancy body mass index (BMI) with plasma glucose concentrations in an oral glucose tolerance test (OGTT) at 12-16 weeks' gestation. A population-based cohort of 1361 pregnant women was recruited in South Karelia, Finland, from March 2013 to December 2016. All participants had their WC, weight, height, HbA1c , and blood pressure measured at 8-14 weeks' gestation and subsequently underwent a 2-h 75-g OGTT, including assessment of fasting insulin concentrations, at 12-16 weeks' gestation. BMI (kg/m2 ) was calculated using self-reported pre-pregnancy weight. Maternal WC ≥80 cm was defined as large. Maternal height ≥166 cm was defined as tall. Data on gestational diabetes treatment was extracted from hospital records. In the total cohort, 901 (66%) of women had an early-pregnancy WC ≥80 cm, which was associated with higher early-pregnancy HbA1c, higherconcentrationsoffasting plasma glucose and serum insulin, higherpost-load plasma glucose concentrations, higherHOMA-IR indices, higher blood pressure levels, and higherfrequencies of pharmacologically treated gestational diabetes, than early-pregnancy WC <80 cm. Maternal height ≥166 cm was negatively associated with 1- and 2-h post-load plasma glucose concentrations. Waist-to-height ratio (WHtR) >0.5 was positively associated with both fasting and post-load plasma glucose concentrations at 12-16 weeks' gestation, even when adjusted for age, smoking, nulliparity, and family history of type 2 diabetes. The best cut-offs for WHtR (0.58 for 1-h plasma glucose, and 0.54 for 2-h plasma glucose) were better predictors of post-load glucose concentrations >90th percentile than the best cut-offs for BMI (28.1kg/m2 for 1-h plasma glucose, and 26.6kg/m2 for 2-h plasma glucose), with areas-under-the-curve (95% confidence interval) 0.73 (0.68-0.79) and 0.73 (0.69-0.77), respectively, for WHtR, and 0.68 (0.63-0.74) and 0.69 (0.65-0.74), respectively, for BMI. In our population-based cohort, early-pregnancy WHtR >0.5 waspositively associatedwith both fasting and post-load glucose concentrations at 12-16 weeks' gestation and performed better than BMI in the prediction of post-load glucose concentrations >90th percentile. Overall, our results underline the importance of evaluating maternal abdominal adiposity in gestational diabetes risk assessment.

The peroxidase-like activity of papain colorimetrically detects H2O2 and glucose with high sensitivity

To date, several glucometer methods are available for patient-guided glycemic control, which diminishes the onset and severity of diabetic complications. In a dipstick urine glucose test, horseradish peroxidase (HRP) oxidizes tetramethylbenzidine (TMB) in the presence of hydrogen peroxide (H2O2) to quantitatively produce a colorimetric signal and predict glucose concentration. Papain, a sulfhydryl protease secreted by the latex of Carica papaya, displays HRP-like activity with high selectivity. However, the current papain-based glucometer method lacks sensitivity. We hypothesized that catalyzing the oxidation of papain under varying buffer conditions (e.g., papain, TMB, and acetic acid concentrations) would improve glucose detection sensitivity. To test this, we performed glucose oxidase (GOx)-mediated glucose oxidation and papain-mediated TMB oxidation under varying sample conditions in vitro and spontaneously coupled them. The aims of this study were to validate the coupling of GOx-mediated glucose oxidation and papain-mediated TMB oxidation and identify the optimized conditions that maximize catalytic activity. We confirmed that glucose oxidase quantitatively produces H2O2 (in a glucose concentration-dependent manner), the presence of which catalyzes papain-mediated production of a colorimetric signal. When two oxidation reactions were coupled, we identified that the sensitivity of papain’s glucose detection increased by ten-fold compared to previous studies with a markedly high correlation (R2 = 0.99). In particular, substituting sodium acetate-acetic acid (HAc-NaAc) buffer with acetic acid and increasing the sample incubation time and papain concentration markedly increased the sensitivity of colorimetric signals. The optimized protocols can be integrated into emerging hydrogel technology to develop a non-invasive glucometer patch predicting blood glucose from the patient’s produced sweat.

Predicting Blood Glucose Concentration after Short-Acting Insulin Injection Using Discontinuous Injection Records.

Diabetes is an increasingly common disease that poses an immense challenge to public health. Hyperglycemia is also a common complication in clinical patients in the intensive care unit, increasing the rate of infection and mortality. The accurate and real-time prediction of blood glucose concentrations after each short-acting insulin injection has great clinical significance and is the basis of all intelligent blood glucose control systems. Most previous prediction methods require long-term continuous blood glucose records from specific patients to train the prediction models, resulting in these methods not being used in clinical practice. In this study, we construct 13 deep neural networks with different architectures to atomically predict blood glucose concentrations after arbitrary independent insulin injections without requiring continuous historical records of any patient. Using our proposed models, the best root mean square error of the prediction results reaches 15.82 mg/dL, and 99.5% of the predictions are clinically acceptable, which is more accurate than previously proposed blood glucose prediction methods. Through the re-validation of the models, we demonstrate the clinical practicability and universal accuracy of our proposed prediction method.

Dual Wavelength based Approach with Partial Least Square Regression for the Prediction of Glucose Concentration

Diabetes mellitus is a group of metabolic disorder characterized by high blood sugar levels. Monitoring of blood glucose levels at regular intervals plays a crucial role in the management of diabetes. The non-invasive real-time monitoring of glucose using near-infrared (NIR), Raman, acoustic and bio-impedance techniques have an edge over available invasive techniques but suffers from low Signal to Noise ratio (S/N) and other interferences. In the present work, we have attempted to improve S/N for the efficient detection of feeble signals corresponding to the physiological glucose concentrations. Investigations were carried out in the NIR region particularly from 800-1400 nm for the identification of the unique absorption features of glucose using UV-Vis NIR spectrophotometer with different ranges of glucose concentrations including 5 g/dl- 45 g/dl, 1400 mg/dl -2500 mg/dl, 35 mg/dl-650 mg/dl . Savitzky Golay (SG) pre-processing filter was applied on the raw data for enhancing the S/N for better prediction of glucose concentrations. The absorption spectra of glucose revealed the presence of a peak at 960 nm. Therefore, considering the absorbance at 960 nm, provided an enhancement in the S/N ratio from 17 dB to 27 dB. Further, partial least square regression (PLSR), has been applied on SG filtered data for a better prediction of glucose concentration with a correlation coefficient ( R 2 ) value of 0.99 and root mean square error of prediction (RMSE) of 2.29 mg/dl. Further, based on the NIR spectral data, we have developed a measurement technique using two LED sources of 950 nm and 860 nm, and a wide detector (700 - 1100 nm) which converts obtained optical signal into voltage. It has been observed that by considering dual wavelength detection points the prediction of glucose concentration is improved. Furthermore, with increase in the test glucose concentrations, the voltage signal decreased corresponding to the 950 nm LED. This is attributed to reduced signal intensities reaching the photodiode as a result of the increase in glucose absorption. Incorporating dual wavelengths for PLSR reduced the RMSE from 8.98 mg/dl to 6.49 mg/dl and also improved the R 2 value from 0.97 to 0.99.

Ensemble Learning Regression Method for Glucose Concentration Prediction System using Colorimetric Paper-based and Smartphones

Prediction of glucose concentration on android smartphones and colorimetric paper-based using the Ensemble learning regression model has been successfully developed. Several successful developments in our research include automatic image segmentation, image correction using the RPCC method, and the development of a regression model for urine glucose predictions. Furthermore, the model was successfully validated for best performance in the respondent's urine susceptible to color change. We used artificial urine at a 0–2000 mg/dl concentration to create a regression model based on Ensemble learning with the boosting optimization method. In addition, we also compared the Ensemble Bagging regression model and the single learner model, Decision Tree. Server-based applications were also developed using RESTful API communication with two servers: an upload server using Node.js and a computing server using the MATLAB Production Server. The testing process results using artificial urine samples showed that the performance of R2 and RRMSE were 0.98 and 0.05 for the Decision Tree and Ensemble Bagging regression models, respectively. While for the Ensemble Boosting regression model, R2 and RRMSE at the testing process are 0.98 and 0.04. The best validation results using respondents' urine samples are shown in the Ensemble Boosting regression model with R2 and RRMSE performance values of 0.97 and 0.06, respectively. The success rate of the application was 100% on both the Samsung Galaxy A51 and Huawei Nova 5T. This research estimated the glucose concentration reasonably well for health monitoring applications.

Prediction of Glucose Concentration in Children with Type 1 Diabetes Using Neural Networks: An Edge Computing Application.

Background: Type 1 Diabetes Mellitus (T1D) is an autoimmune disease that can cause serious complications that can be avoided by preventing the glycemic levels from exceeding the physiological range. Straightforwardly, many data-driven models were developed to forecast future glycemic levels and to allow patients to avoid adverse events. Most models are tuned on data of adult patients, whereas the prediction of glycemic levels of pediatric patients has been rarely investigated, as they represent the most challenging T1D population. Methods: A Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) Recurrent Neural Network were optimized on glucose, insulin, and meal data of 10 virtual pediatric patients. The trained models were then implemented on two edge-computing boards to evaluate the feasibility of an edge system for glucose forecasting in terms of prediction accuracy and inference time. Results: The LSTM model achieved the best numeric and clinical accuracy when tested in the .tflite format, whereas the CNN achieved the best clinical accuracy in uint8. The inference time for each prediction was far under the limit represented by the sampling period. Conclusion: Both models effectively predict glucose in pediatric patients in terms of numerical and clinical accuracy. The edge implementation did not show a significant performance decrease, and the inference time was largely adequate for a real-time application.

Application of Two-Dimensional Fluorescence Spectroscopy for the On-Line Monitoring of Teff-Based Substrate Fermentation Inoculated with Certain Probiotic Bacteria.

There is increasing demand for cereal-based probiotic fermented beverages as an alternative to dairy-based products due to their limitations. However, analyzing and monitoring the fermentation process is usually time consuming, costly, and labor intensive. This research therefore aims to apply two-dimensional (2D)-fluorescence spectroscopy coupled with partial least-squares regression (PLSR) and artificial neural networks (ANN) for the on-line quantitative analysis of cell growth and concentrations of lactic acid and glucose during the fermentation of a teff-based substrate. This substrate was inoculated with mixed strains of Lactiplantibacillus plantarum A6 (LPA6) and Lacticaseibacillus rhamnosus GG (LCGG). The fermentation was performed under two different conditions: condition 1 (7 g/100 mL substrate inoculated with 6 log cfu/mL) and condition 2 (4 g/100 mL substrate inoculated with 6 log cfu/mL). For the prediction of LPA6 and LCGG cell growth, the relative root mean square error of prediction (pRMSEP) was measured between 2.5 and 4.5%. The highest pRMSEP (4.5%) was observed for the prediction of LPA6 cell growth under condition 2 using ANN, but the lowest pRMSEP (2.5%) was observed for the prediction of LCGG cell growth under condition 1 with ANN. A slightly more accurate prediction was found with ANN under condition 1. However, under condition 2, a superior prediction was observed with PLSR as compared to ANN. Moreover, for the prediction of lactic acid concentration, the observed values of pRMSEP were 7.6 and 7.7% using PLSR and ANN, respectively. The highest error rates of 13 and 14% were observed for the prediction of glucose concentration using PLSR and ANN, respectively. Most of the predicted values had a coefficient of determination (R2) of more than 0.85. In conclusion, a 2D-fluorescence spectroscopy combined with PLSR and ANN can be used to accurately monitor LPA6 and LCGG cell counts and lactic acid concentration in the fermentation process of a teff-based substrate. The prediction of glucose concentration, however, showed a rather high error rate.

Avoiding misleading predictions in fluorescence-based soft sensors using autoencoders

Excitation-emission matrix (EEM) fluorescence data can be explored by using deep convolutional neural networks to enhance the predictive performance of bioprocess variables. This article proposes the use of residual neural networks (ResNet) for the prediction of ethanol, glucose, and biomass concentrations of S. cerevisiae cultivations based on fluorescence data collected in situ. A trust screening for unkown samples, based on autoencoder (AE) reconstruction error, is also proposed. Its characteristic of reconstructing the inputs is the key feature to avoid misleading predictions and forecast if a new sample should be trusted as usual or flagged as abnormal. The 83 layers deep ResNet successfully predicted the desired outputs, with R2 higher than 0.98, in the test subset. The best-fitted autoencoder had a 3-layer architecture, with three neurons in the bottleneck and using rectified linear unit (ReLU) activation for the encoder and linear activation for the decoder. The mean reconstruction Root Mean Square Error (RMSE) for the fermentation's EEMs was 4.61 rel. fluorescence intensity units, representing an error smaller than 1% (of the total amplitude of change). To evaluate the AE capability to work as trust screening, random fluorescence intensity was added to the Ex450/Em530 fluorescence pair (related to flavins) in some samples, creating a defective dataset. The dataset was evaluated with the trained AE and the ResNet model to compare reconstruction errors and bioprocess concentrations. The AE was able to identify the samples with added errors, and, as expected, the defective samples also presented higher predictive errors in general. The higher the AE's reconstruction RMSE, the less the new sample should be trusted to avoid misleading predictions.

Study on noninvasive blood glucose detection method using the near-infrared light based on particle swarm optimization and back propagation neural network

Most of the existing near-infrared noninvasive blood glucose detection models focus on the relationship between near-infrared absorbance and blood glucose concentration, but do not consider the impact of human physiological state on blood glucose concentration. In order to improve the performance of prediction model, particle swarm optimization (PSO) algorithm was used to train the structure paramters of back propagation (BP) neural network. Moreover, systolic blood pressure, pulse rate, body temperature and 1 550 nm absorbance were introduced as input variables of blood glucose concentration prediction model, and BP neural network was used as prediction model. In order to solve the problem that traditional BP neural network is easy to fall into local optimization, a hybrid model based on PSO-BP was introduced in this paper. The results showed that the prediction effect of PSO-BP model was better than that of traditional BP neural network. The prediction root mean square error and correlation coefficient of ten-fold cross-validation were 0.95 mmol/L and 0.74, respectively. The Clarke error grid analysis results showed that the proportion of model prediction results falling into region A was 84.39%, and the proportion falling into region B was 15.61%, which met the clinical requirements. The model can quickly measure the blood glucose concentration of the subject, and has relatively high accuracy.

Development of non-invasive blood glucose regression based on near-infrared spectroscopy combined with a deep-learning method

Extracting micro-scale spectral features from dynamic blood glucose concentrations is extremely difficult when using non-invasive measurement methods. This work proposes a new machine-learning method based on near-infrared spectroscopy, a deep belief network (DBN), and a support vector machine to improve prediction accuracy. First, the standard oral glucose tolerance test was used to collect near-infrared spectroscopy and actual blood glucose concentration values for specific wavelengths (1200, 1300, 1350, 1450, 1600, 1610, and 1650 nm); the blood glucose concentrations were within a clinical range of 70 ∼ 220 mg dl−1. Second, based on the DBN model, high-dimensional deep features of the non-invasive blood glucose spectrum were extracted. These were used to establish a support vector regression (SVR) model and to quantitatively analyze the influence of the spectral sample size and corresponding feature dimensions (i.e. DBN structure) on prediction accuracy. Finally, based on data from six volunteers, a comparative analysis of the SVR model’s prediction accuracy was performed both before and after using high-dimensional deep features. For volunteer 1, when the DBN-based high-dimensional deep features were used, the root mean square error of the SVR model was reduced by 71.67%, and the correlation coefficient (R 2) and the P value of the Clark grid analysis (P) were increased by 13.99% and 6.28%, respectively. Moreover, we had similar results when the proposed method was carried out on the data of other volunteers. The results show that the presented algorithm can play an important role in dynamic non-invasive blood glucose concentration prediction and can effectively improve the accuracy of the SVR model. Further, by applying the algorithm to six independent sets of data, this research also illustrates the high-precision regression and generalization capabilities of the DBN-SVR algorithm.

Examining Type 1 Diabetes Mathematical Models Using Experimental Data

Type 1 diabetes requires treatment with insulin injections and monitoring glucose levels in affected individuals. We explored the utility of two mathematical models in predicting glucose concentration levels in type 1 diabetic mice and determined disease pathways. We adapted two mathematical models, one with -cells and the other with no -cell component to determine their capability in predicting glucose concentration and determine type 1 diabetes pathways using published glucose concentration data for four groups of experimental mice. The groups of mice were numbered Mice Group 1–4, depending on the diabetes severity of each group, with severity increasing from group 1–4. A Markov Chain Monte Carlo method based on a Bayesian framework was used to fit the model to determine the best model structure. Akaike information criteria (AIC) and Bayesian information criteria (BIC) approaches were used to assess the best model structure for type 1 diabetes. In fitting the model with no -cells to glucose level data, we varied insulin absorption rate and insulin clearance rate. However, the model with -cells required more parameters to match the data and we fitted the -cell glucose tolerance factor, whole body insulin clearance rate, glucose production rate, and glucose clearance rate. Fitting the models to the blood glucose concentration level gave the least difference in AIC of , and a difference in BIC of for Mice Group 4. The estimated AIC and BIC values were highest for Mice Group 1 than all other mice groups. The models gave substantial differences in AIC and BIC values for Mice Groups 1–3 ranging from to . Our results suggest that the model without -cells provides a more suitable structure for modelling type 1 diabetes and predicting blood glucose concentration for hypoglycaemic episodes.

Blood glucose concentration prediction based on VMD-KELM-AdaBoost.

The time series of blood glucose concentration in diabetic patients are time-varying, nonlinear, and non-stationary. In order to improve the accuracy of blood glucose prediction, a multi-scale combination short-term blood glucose prediction model was constructed by combining the variational mode decomposition (VMD) method, the kernel extreme learning machine (KELM), and the AdaBoost algorithm (VMD-ELM-AdaBoost). Firstly, the blood glucose concentration series were decomposed into a set of intrinsic modal functions (IMFs) with different scales by the VMD method. On this basis, the KELM neural network and AdaBoost algorithm are combined to predict each IMF component. Finally, the cumulative blood glucose concentration prediction value is obtained by accumulating the KELM-AdaBoost prediction results of each IMF. The time series of measured blood glucose concentration were used for experimental analysis; the experimental results show that the proposed VMD-KELM-AdaBoost method has higher prediction accuracy compared with the classical prediction models such as ELM, KELM, SVM, and LSTM. The proposed VMD-KELM-AdaBoost model can still achieve high prediction accuracy 60 min in advance (the mean values of RMSE, MAPE, and CC are about 10.1422, 4.8629%, and 0.8737 respectively); in Clarke error mesh analysis, the proportion of falling into A region is about 95.7%; the sensitivity and false alarm rate of early alarm of hypoglycemia were 94.8% and 7.7%, respectively. Graphical abstract We have proposed a new prediction model. In the first part, for reducing thenon-stationarity, the data of blood glucose concentration was decomposed as a series ofIMF by VMD. In the second part, a prediction model based KELM and Adaboost wasestablished. In the third part, the KELM-Adaboost model was used to predict each IMF,and the predicted values of all IMFS were superimposed to obtain the final predictionresult of blood glucose concentration.

AI powered electrochemical multi-component detection of insulin and glucose in serum

Multi-component detection of insulin and glucose in serum is of great importance and urgently needed in clinical diagnosis and treatment due to its economy and practicability. However, insulin and glucose can hardly be determined by traditional electrochemical detection methods. Their mixed oxidation currents and rare involvement in the reaction process make it difficult to decouple them. In this study, AI algorithms are introduced to power the electrochemical method to conquer this problem. First, the current curves of insulin, glucose, and their mixed solution are obtained using cyclic voltammetry. Then, seven features of the cyclic voltammetry curve are extracted as characteristic values for detecting the concentrations of insulin and glucose. Finally, after training using machine learning algorithms, insulin and glucose concentrations are decoupled and regressed accurately. The entire detection process only takes three minutes. It can detect insulin at the pmol level and glucose at the mmol level, which meets the basic clinical requirements. The average relative error in predicting insulin concentrations is around 6.515%, and that in predicting glucose concentrations is around 4.36%. To verify the performance and effectiveness of the proposed method, it is used to determine the concentrations of insulin and glucose in fetal bovine serum and real clinical serum samples. The results are satisfactory, demonstrating that the method can meet basic clinical needs. This multi-component testing system delivers acceptable detect limit and accuracy and has the merits of low cost and high efficiency, holding great potential for use in clinical diagnosis.