Noninvasive Diabetes Research Articles

Machine-learning-based diabetes classification method using blood flow oscillations and Pearson correlation analysis of feature importance

Abstract Diabetes is a global health issue affecting millions of people and is related to high morbidity and mortality rates. Current diagnostic methods are primarily invasive, involving blood sampling, which can lead to infection and increased patient stress. As a result, there is a growing need for noninvasive diabetes diagnostic methods that are both accurate and fast. High measurement accuracy and fast measurement time are essential for effective noninvasive diabetes diagnosis; these can be achieved using diffuse speckle contrast analysis (DSCA) systems and artificial intelligence algorithms. In this study, we use a machine learning algorithm to analyze rat blood flow signals measured using a DSCA system with simple operation, easy fabrication, and fast measurement for helping diagnose diabetes. The results confirmed that the machine learning algorithm for analyzing blood flow oscillation data shows good potential for diabetes classification. Furthermore, analyzing the blood flow reactivity test revealed that blood flow signals can be quickly measured for diabetes classification. Finally, we evaluated the influence of each blood flow oscillation data on diabetes classification through feature importance and Pearson correlation analysis. The results of this study should provide a basis for the future development of hemodynamic-based disease diagnostic methods.

Enhanced Diabetes Detection and Blood Glucose Prediction Using TinyML-Integrated E-Nose and Breath Analysis: A Novel Approach Combining Synthetic and Real-World Data

Diabetes mellitus, a chronic condition affecting millions worldwide, necessitates continuous monitoring of blood glucose level (BGL). The increasing prevalence of diabetes has driven the development of non-invasive methods, such as electronic noses (e-noses), for analyzing exhaled breath and detecting biomarkers in volatile organic compounds (VOCs). Effective machine learning models require extensive patient data to ensure accurate BGL predictions, but previous studies have been limited by small sample sizes. This study addresses this limitation by employing conditional generative adversarial networks (CTGAN) to generate synthetic data from real-world tests involving 29 healthy and 29 diabetic participants, resulting in over 14,000 new synthetic samples. These data were used to validate machine learning models for diabetes detection and BGL prediction, integrated into a Tiny Machine Learning (TinyML) e-nose system for real-time analysis. The proposed models achieved an 86% accuracy in BGL identification using LightGBM (Light Gradient Boosting Machine) and a 94.14% accuracy in diabetes detection using Random Forest. These results demonstrate the efficacy of enhancing machine learning models with both real and synthetic data, particularly in non-invasive systems integrating e-noses with TinyML. This study signifies a major advancement in non-invasive diabetes monitoring, underscoring the transformative potential of TinyML-powered e-nose systems in healthcare applications.

Pilot study for non-invasive diabetes detection through classification of photoplethysmography signals using convolutional neural networks

Diabetes is a chronic disorder affecting vascular health, often altering pulse wave characteristics. Traditional pulse wave analysis (PWA) methods face challenges such as variability and complexity of signals. This study aims to overcome these limitations by leveraging deep learning models for more accurate and efficient classification. The methodology used in this study involves four key steps: data collection, data preprocessing, Convolutional Neural Network (CNN) model development, and model evaluation. Primary data were collected using a multipara patient monitor, including finger photoplethysmography (PPG) signals, blood pressure, mean arterial pressure, oxygen saturation, and pulse rate. Single pulse wave cycles from 60 healthy individuals and 60 patients with type 2 diabetes underwent preprocessing. The CNN model was trained using 50 PPG images from each group and achieved a training accuracy of 92%. The prediction capability of the model was evaluated using 20 unseen images, comprising 10 healthy and 10 diabetes PPG images. It attained a 90% overall test accuracy in distinguishing between PPG images of individuals with diabetes and those who are healthy. These findings suggest that CNN-based analysis of PPG signals provides a precise, non-invasive tool for diabetes screening. To further enhance accuracy, future studies should focus on increasing the dataset size and performing hyperparameter tuning to optimize the CNN model.

Noninvasive Monitoring of Glycemia Level in Diabetic Patients by Wearable Advanced Biosensors.

We report on the possibility of noninvasive diabetes monitoring through continuous analysis of sweat. The prediction of the blood glucose level in diabetic patients is possible on the basis of their sweat glucose content due to the positive correlation discovered. The ratio between the blood glucose and sweat glucose concentrations for a certain diabetic subject is stable within weeks, excluding requirements for frequent blood probing. The glucose variations in sweat display allometric (non-linear) dependence on those in blood, allowing more precise blood glucose estimation. Selective (avoiding false-positive responses) and sensitive (sweat glucose is on average 30-50 times lower) detection is possible with biosensors based on the glucose oxidase enzyme coupled with a Prussian Blue transducer. Reliable glucose detection in just secreted sweat would allow noninvasive monitoring of the glycemia level in diabetic patients.

Hybrid CNN-LSTM for Predicting Diabetes: A Review.

Diabetes is a common and deadly chronic disease caused by high blood glucose levels that can cause heart problems, neurological damage, and other illnesses. Through the early detection of diabetes, patients can live healthier lives. Many machine learning and deep learning techniques have been applied for noninvasive diabetes prediction. The results of some studies have shown that the CNN-LSTM method, a combination of CNN and LSTM, has good performance for predicting diabetes compared to other deep learning methods. This paper reviews CNN-LSTM-based studies for diabetes prediction. In the CNNLSTM model, the CNN includes convolution and max pooling layers and is applied for feature extraction. The output of the max-pooling layer was fed into the LSTM layer for classification. The CNN-LSTM model performed well in extracting hidden features and correlations between physiological variables. Thus, it can be used to predict diabetes. The CNNLSTM model, like other deep neural network architectures, faces challenges such as training on large datasets and biological factors. Using large datasets can further improve the accuracy of detection. The CNN-LSTM model is a promising method for diabetes prediction, and compared with other deep-learning models, it is a reliable method.

Analysis of Non-Invasive Diabetes Detection Technology

Analysis of Non-Invasive Diabetes Detection Technology

One step synthesis of Rh embedded hollow spherical WO₃ composite for enhanced acetone sensing: A promising approach for non-invasive diabetes monitoring

This study investigated the enhancement of WO₃/Rh composite gas sensors for improved acetone detection. WO₃/Rh composite samples (WO₃/Rh-1, WO₃/Rh-2, WO₃/Rh-3) were synthesized via spray pyrolysis, and materials were studied using XRD, SEM, and TEM. These analyses confirmed successful introduction of Rh nanoparticles and indicated favorable alterations in structure and morphology. Gas sensing tests with varying acetone concentrations (0.1–4 PPM, balanced with simulated exhaled gas with RH% of ∼30 %) demonstrated that Rh introduction significantly improved sensitivity and selectivity. Notably, WO₃/Rh-3 exhibited the highest performance, with enhanced response times and selectivity critical for diabetes monitoring. These results highlight the potential of WO₃/Rh sensors in medical diagnostics, offering a promising approach for the early detection of diabetes through breath analysis.

Enhanced acetone detection for non-invasive diabetes monitoring by atomic layer deposited WO3 nanoparticle on hierarchical In2O3 particles

This study addressed the critical need for non-invasive monitoring of diabetes by proposing an acetone gas sensor based on hierarchical In2O3 with atomic layer deposition (ALD)-deposited WO3 nanoparticles. The sensor fabrication involved a carefully designed process, leveraging ALD to control WO3 deposition, ensuring uniform distribution, and mitigating agglomeration. The resulting composite exhibited enhanced sensitivity, making it promising for detecting acetone, a key biomarker for diabetes. Material synthesis, including hydrothermal formation of In2O3 hierarchy particles and ALD of WO3, was meticulously conducted. Comprehensive characterizations, involving SEM, TEM, EDX, XRD, XPS, and BET, validated the successful synthesis and deposition. The sensor’s response to varying acetone concentrations (50–2000 ppb) was systematically investigated, revealing a positive correlation. The In2O3/WO3–2 sensor exhibited the highest sensitivity, attributed to the catalytic properties of WO3. The proposed sensor presented a cost-effective, sensitive, and selective solution, paving the way for non-invasive diabetes monitoring.

Noninvasive Diabetes Detection through Human Breath Using TinyML-Powered E-Nose.

Volatile organic compounds (VOCs) in exhaled human breath serve as pivotal biomarkers for disease identification and medical diagnostics. In the context of diabetes mellitus, the noninvasive detection of acetone, a primary biomarker using electronic noses (e-noses), has gained significant attention. However, employing e-noses requires pre-trained algorithms for precise diabetes detection, often requiring a computer with a programming environment to classify newly acquired data. This study focuses on the development of an embedded system integrating Tiny Machine Learning (TinyML) and an e-nose equipped with Metal Oxide Semiconductor (MOS) sensors for real-time diabetes detection. The study encompassed 44 individuals, comprising 22 healthy individuals and 22 diagnosed with various types of diabetes mellitus. Test results highlight the XGBoost Machine Learning algorithm's achievement of 95% detection accuracy. Additionally, the integration of deep learning algorithms, particularly deep neural networks (DNNs) and one-dimensional convolutional neural network (1D-CNN), yielded a detection efficacy of 94.44%. These outcomes underscore the potency of combining e-noses with TinyML in embedded systems, offering a noninvasive approach for diabetes mellitus detection.

Enhanced prediction of abnormal glucose tolerance using an extended non-invasive risk score incorporating routine renal biochemistry

IntroductionType 2 diabetes is preventable in subjects with impaired glucose tolerance based on 2-hour plasma glucose (2hPG) during 75 g oral glucose tolerance test (OGTT). We incorporated routine biochemistry to...

Unlocking Diabetic Acetone Vapor Detection by A Portable Metal-Organic Framework-Based Turn-On Optical Sensor Device.

Despite exhaled human breath having enabled noninvasive diabetes diagnosis, selective acetone vapor detection by fluorescence approach in the diabetic range (1.8-3.5ppm) remains a long-standing challenge. A set of water-resistant luminescent metal-organic framework (MOF)-based composites have been reported for detecting acetone vapor in the diabetic range with a limit of detection of 200 ppb. The luminescent materials possess the ability to selectively detect acetone vapor from a mixture comprising nitrogen, oxygen, carbon dioxide, water vapor, and alcohol vapor, which are prevalent in exhaled breath. It is noteworthy that this is the first luminescent MOF material capable of selectively detecting acetone vapor in the diabetic range via a turn-on mechanism. The material can be reused within a matter of minutes under ambient conditions. Industrially pertinent electrospun luminescent fibers are likewise fabricated alongside various luminescent films for selective detection of ultratrace quantities of acetone vapor present in the air. Ab initio theoretical calculations combined with in situ synchrotron-based dosing studies uncovered the material's remarkable hypersensitivity toward acetone vapor. Finally, a freshly designed prototype fluorescence-based portable optical sensor is utilized as a proof-of-concept for the rapid detection of acetone vapor within the diabetic range.

NanoLC-timsTOF-Assisted Analysis of Glycated Albumin in Diabetes-Affected Plasma and Tears.

Glycation is a spontaneous and nonenzymatic glycosylation. Glycated albumin (GA), which serves as an important biomarker in plasma in the diagnosis and characterization of diabetes, can be passively filtered from the plasma to tears. Tears are important targets for research in clinical diagnostics due to the ability to collect this biofluid noninvasively and repeatably. Therefore, the analysis of GA in tear film provides information for monitoring diabetes progression independent of blood pathologies. Due to the limited volume (1-5 μL) of natural tear film, we developed a small volume assay using a nano liquid chromatography-trapped ion mobility spectrometry-time-of-flight MS (nanoLC-timsTOF) platform for the analysis of glycated albumin in human plasma and tear films affected by diabetes. The peptides containing lysine 525, which is the main glycation site in GA, were relatively quantified and represented as the GA level. The results of the measurements showed that GA levels were significantly higher in diabetes-affected plasma and tears compared to controls with a p-value < 0.01. A strong correlation of glycated albumin levels was observed for the plasma and tear film in diabetes samples (Pearson coefficient 0.92 with a p-value 0.0012). Moreover, the number of GA glycation sites was significantly higher in diabetes-affected plasma and tear comparatively to controls. Among all the glycation sites in plasma albumin, the GA level quantified by lysine 136/137 had a strong correlation with more commonly used lysine 525, suggesting that lysine 136 /137 is an alternative diabetes biomarker in plasma. Overall, our findings demonstrate GA in tears as a biomarker for monitoring diabetes progression, highlighting new possibilities for quick and noninvasive diabetes detection and monitoring.

Binary fire hawks optimizer with deep learning driven non-invasive diabetes detection and classification.

Non-invasive diabetes detection refers to the utilization and development of technologies and methods that can monitor and diagnose diabetes without requiring invasive procedures, namely invasive glucose monitoring or blood sampling. The objective is to provide a more convenient and less burdensome approach to screening and management of diabetes. It is noteworthy that while non-invasive method offers promising avenues for diabetes detection, they frequently require validation through clinical studies and might have limitation in terms of reliability and accuracy than classical invasive approaches. In recent times, deep learning (DL) and feature selection (FS) are used to monitor and diagnose diabetes accurately without requiring invasive procedures. This technique combines the FS method with the DL algorithm for making accurate predictions and extracting relevant features from non-invasive data. This article introduces a new Binary Fire Hawks Optimizer with Deep Learning-Driven Non-Invasive Diabetes Detection and Classification (BFHODL-NIDDC) technique. The major intention of the BFHODL-NIDDC technique focuses on the involvement of non-invasive procedures for the detection of diabetes. In the BFHODL-NIDDC technique, data preprocessing is initially performed to preprocess the input data. Next, the BFHO algorithm chooses an optimal subset of features and improves the classifier results. For the identification of diabetes, multichannel convolutional bidirectional long short-term memory (MC-BLSTM) model is used. At last, the beetle antenna search (BAS) algorithm is used for the hyperparameter selection of the MC-BLSTM method which in turn enhances the detection performance of the MC-BLSTM model. A series of simulations were conducted on the diabetes dataset to assess the diabetes detection performance of the BFHODL-NIDDC technique. The experimental outcomes illustrated better performance of the BFHODL-NIDDC method over other recent approaches in terms of different metrics (Tab. 4, Fig. 9, Ref. 23). Keywords: diabetes, non-invasive detection, binary fire hawks optimizer, deep learning, hyperparameter tuning.

Non-invasive Diabetes Mellitus Diagnostics Using Laser-Induced Breakdown Spectroscopy and Support Vector Machine Algorithm

Non-invasive Diabetes Mellitus Diagnostics Using Laser-Induced Breakdown Spectroscopy and Support Vector Machine Algorithm

Polyazine nanoparticles as anchors of PQQ glucose dehydrogenase for its most efficient bioelectrocatalysis

We report on the drop-cast production of glucose biosensors based on the most efficient bioelectrocatalysis by pyrroloquinoline quinone dependent glucose dehydrogenase (PQQ GDH). To orient the enzyme upon immobilization we suggest using poly(Methylene Green) (p(MG)) nanoparticles acting as anchors. Synthesis of polymeric anchors has been carried out in course of Methylene Green electropolymerization, which allowed to tune polymer-to-monomer ratio in the drop-cast mixtures varying the monomer concentration and applying different number of potential sweep cycles. Except for elimination of electrochemical step for the electrode modification, opening the prospects for mass-production, the use of drop-cast enzyme anchors provides advantageous characteristics of the resulting biosensors. Catalytic current of PQQ GDH immobilized over p(MG) nanoparticles obtained in optimal conditions is increased only 2–2.5 times after addition of the freely diffusing mediator. Obviously, the lowest ratio of mediated-to-reagentless current points to the most efficient bioelectrocatalysis. The obtained ratio is 2.5 times lower than that for biosensors based on electropolymerized p(MG) films and practically an order of magnitude lower than that for the best reagentless sensors based on PQQ GDH immobilized over conductive nanomaterials. The achieved most efficient bioelectrocatalysis provides high sensitivity of the elaborated biosensors even at 0.0 V potential, which allows to operate them in power generation mode and control relative sweat glucose variation as a tool for non-invasive diabetes monitoring.

Black phosphorus nanosheets-sensitized Zn-doped α-Fe2O3 nanoclusters for trace acetone detection

Acetone vapor not only poses a significant threat to the ecological environment and human health but serves as a specific volatile organic compounds (VOCs) biomarker within the exhaled breath to distinguish individual diabetes, thereby urgently necessitating a sensitive and selective acetone detection. To this end, conductometric gas sensors featuring black phosphorus nanosheets (BP)-decorated Zn-doped hematite (α-Fe2O3) nanoclusters as the sensitive material were prepared. By subtly modulating the componential ratio and operation temperature, the optimal Zn-doped α-Fe2O3/BP sensor delivered a prominent response of 5.32 and short response/recovery times of 8/68 s toward 0.5 ppm acetone at 160 °C as well as a large sensitivity of 5.3/ppm over the concentration range of 0.1–1.5 ppm. Moreover, favorable selectivity, repeatability, and long-term stability were demonstrated. The superior sensing performance after incorporating a tiny but moderate amount of BP over Zn-doped α-Fe2O3 sensors could be attributed to the consequent enlarged specific surface area, increased oxygen vacancies, and numerous p-n heterojunctions. Furthermore, additional polylactic acid (PLA) encapsulation membrane endowed the as-prepared sensors with a remarkably improved humidity tolerance, showcasing a huge application potential in non-invasive diabetes diagnosis from exhaled breath featuring high-humidity conditions in the future.

Zigzagging graphitic ribbons: Synthesis, characterization, and acetone vapor sensing

The fast sensing highly volatile and flammable liquids of organic compounds is an urgent scientific goal for developing devices aimed at health and safety monitoring. The high superficial area, high porosity, and electrochemical properties of carbon nanotube/fiber sponges make them promising materials for sensing organic compounds. In this study, we synthesized zigzagging graphitic ribbons (ZGR) sponges employing the aerosol-assisted chemical vapor deposition (AACVD) method at temperatures of 980 °C, 1020 °C, and 1050 °C. We characterized the morphology, composition, thermal stability, and surface chemistry of the ZGRs using various techniques. The ZGRs exhibited a defective, corrugated, and oxygen-functionalized surface. XPS characterizations revealed the presence of ether, carbonyl, carboxyl, aldehyde, and hydroxyl groups on the ZGR surface. TEM characterizations showed graphitic materials with unusual black zones attributed to sp3 carbon hybridization or strained graphite. Sensors based on ZGR sponges were used at room temperature for detecting acetone gas vapor. The electrical resistance response was monitored by varying the concentration of acetone gas vapor. The results revealed that the sensors remained stable over multiple cycles and exhibited fast response times (2–12 s). Interestingly, acetone serves as a selective breath marker for diabetes, suggesting that our sensors could be used for non-invasive diabetes monitoring. Finally, we performed van der Waals density functional theory calculations to elucidate the TEM black zones and the interaction of the acetone molecule with oxygen functional groups hosted in graphitic materials.

HI-Shaped Antenna for Non-Invasive Diabetes Measurement and Monitor Fluctuating Diabetes

The HI antenna senses the human pancreas dielectric radiation for diabetic measurement. Existing passive sensor antennas sense the dielectric radiation from the pancreas region at frequencies of 402.5 MHz and 2.4 GHz for a relative permittivity of 61.2155 and 57.201, respectively. The proposed antenna senses the dielectric properties of diabetic-affected pancreas, such as low and high-fat diabetic pancreas, in the frequency range between 1.5 GHz and 3 GHz. The relative permittivity of the diabetic pancreas is in the range of 48.235 to 65.508. The proposed antenna can sense the diabetic range between 70 mg/dl and 475 mg/dl based on a change in dB level, whereas the existing diabetic sensing antenna measures the diabetic level based on resonance frequency. The resonance-frequency-based diabetic measurement shows inaccurate results. The HI antenna senses pancreas radiation effectively because of the shape and size of its slot, which covers the pancreas region of the human body without generating noise due to the crumbling effect during pancreas dielectric radiation acquisition. The proposed HI-shaped antenna is mounted in different parts of the human body, such as the hand, finger, stomach, and pancreas, for measuring fluctuating diabetes. Based on the experimental results of the proposed HI-shaped antenna, the pancreas is in an optimal location among the various parts of the body. The proposed HI-shaped antenna-based dielectric signal statical values were correlated with diabetic laboratory values for prediction of the diabetic value. The proposed antenna measures the fluctuation in diabetes with 85% accuracy

Highly tunable and ultrasensitive hybrid plasmonic platform for multi-spectral sensing: Non-invasive diabetes surveillance and global warming directives

The development of tunable and extremely sensitive devices, as well as the emergence of metamaterials, has tremendously aided the evolution of plasmonic sensing technologies. However, the natural property of metamaterials severely restricts the scalability and application potential of mono-functional sensor systems. To address this shortcoming, we have investigated an ultranarrow five-band plasmonic sensor with structural tunability and high sensing behaviors in compatibility with multi-purpose application scenarios, using the finite-difference time-domain approach rigorously. Meanwhile, we theoretically formulated a hybridization model for the mutual induction of Fabry–Pérot resonances and localized surface plasmon polariton resonance, providing a theoretical basis for the utilization and conception of the highly-sensitive tunable plasmonic sensor. Most notably, we discovered and interpreted far-field phase extinction inducing separation of polarization resonance peaks in physical mechanisms. The tunability of its mutual inductance effect at the intersection distances from two-intersecting silver nanogroove was adequately exploited to further validate and effectively attain five ultra-narrow absorptivities of 96.70%, 93.93%, 99.73%, 100.00%, and 89.72% by top-level optimization. In addition, the redundant proximity between nearby peaks is compensated, extending the operating band and enhancing the practicability in multi-purpose scenarios. As a reliable source for future optical sensing, it performs well in the visible to near-infrared region, enabling detection response to non-invasive diabetes mellitus, global climate warming, and unrestricted temperature and pressure measurement. Meanwhile, minimum FWHM = 4.806 nm and maximum FOM = 115.13 RIU−1 are obtained at 293.15 K, with an associated highest sensitivity of 557.93 nm/RIU. Moreover, the fabrication technologies development of electron beam lithography and focused ion beam etching technologies enables large-scale mass production at low cost and long-term stability.

Pulse Power Generation Chronoamperometry as an Advanced Readout for (Bio)sensors: Application for Noninvasive Diabetes Monitoring.

We propose pulse power generation (PPG) amperometry as an advanced readout realized for Prussian blue (PB)-based (bio)sensors. In contrast to the conventional power generation mode, when the current response is generated upon continuous short-circuiting, the suggested pulse regime is fulfilled by periodic opening and shorting of the circuit. Despite PB being electroactive, the pulse readout is advantageous over conventional steady-state power generation, providing up to a 15-fold increased signal-to-background ratio as well as dramatically improved sensitivity exceeding 10 A·M-1·cm-2 for H2O2 sensors and 3.9 A·M-1·cm-2 for glucose biosensors. Such analytical performance characteristics are, most probably, achieved due to the enrichment of the diffusion layer by analyte mass transfer from the bulk upon opening of the circuit. Due to an improved sensitivity-to-background ratio, reduced flow-rate dependence, and enhanced operational stability, the regime allows reliable monitoring of blood glucose variations through sweat analysis with the on-skin device.